can still play their favorite sports. 17 Educational. Tourist .... that âreading other travelersâ online reviews increases confidence in decisions, makes it easier to ...
DESIGN AND DEVELOPMENT OF HYBRID RECOMMENDER SYSTEM FOR TOURISM
Saso Koceski Biljana Petrevska
1
CONTENT 1. Introduction ……………………………………………………………… 3 2. Problems and challenges in tourism …………………………………….. 6 2.1. Past patterns in tourism development: world, regional and national perspective …………………………………..………………. 7 2.2. Current status on tourism development ………………..…………… 14 2.3. Future challenges ………………………………………..………….. 16 2.4. ICT and e-tourism …………………………………………..………. 19 3. Examples in tourism recommendation systems …………………...…… 21 3.1. Introduction to recommendation systems ……..…………..………... 21 3.2. Terms and concepts …………………………………………...…..… 22 3.3. Recommendation process ……………………………………..….… 23 3.3.1. Information recollection …………………………..……… 25 3.3.2. Selection ……………………………………………..…… 25 3.3.3. Transformation …………………………………….……... 26 3.3.4. Structuring …………………………...……………..…….. 26 3.3.5. Presentation ………………………...……………….……. 26 3.3.6. Feedback ………………………………………………..... 26 3.4. Tourists‟ preferences and related work …………………………..…. 27 3.5. Tourism recommendation systems and related work ……………...... 28 4. Recommendation algorithm ……………………………………...…….. 29 5. Design and implementation of recommendation system …………....…. 37 5.1. System architecture …………………………………………..……... 37 5.2. Web based graphical user interface …………………………..….…. 38 5.3. Implementation issues ……………………………………..………... 41 6. Empirical evidence on Macedonia …………………………………...… 44 7. Necessity of developing tourism recommender in Macedonia ………… 54 8. Conclusion and future work ……………………………………………. 56 9. References ………………………………………………………...……. 58 2
1. INTRODUCTION Everyone identifies tourism industry as a source for generating numerous positive impacts. Generally, tourism contributes to economic growth and development, promoting global community, international understanding and peace, providing tourism and recreational facilities to local people, improving living standards, stimulating local commerce and industry, reinforcing the preservation of heritage and tradition and so forth. Moreover tourism can contribute for integrating less developed regions or giving them equal access to the fruits of growth. In this respect, one of the major challenges consists of setting up mechanisms to improve competitiveness and quality of tourism at regional and local levels, as well as to ensure sustainable and balanced tourism development at national levels. At the same time, tourism has emerged as a major factor for regional economic development. Regardless the nature, tourism has a major economic and social impact at regional and local levels in the areas where tourism activities take place. So, some regions were highly positively influenced by tourism impacts, like mainly coastal (EmiliaRomagna in Italy), mountainous (Valais in Switzerland), urban and historic (Ile-deFrance in France) or regions with exceptional natural resources (Quebec in Canada, Arizona in the United States). Additionally, regions with different profiles can also benefit from the growth of tourism. In this line, they can be rural, promoting green tourism, leisure and nature activities (Queensland in Australia), very remote, (Greenland in Denmark) or regions undergoing industrial restructuring (Nord-Pas-deCalais in France). The regional development of tourism can trigger general economic growth by creating a new dynamics. It can also contribute to better land use planning by countering rapid urbanisation in developed countries and by attracting populations to new regions where tourism is developing. However, some guidelines for development must be laid down in order to preserve resources, ensure complementarity between areas and define tourism poles (which may not coincide with administrative boundaries). Yet, tourism development in the underdeveloped areas enables development of the periphery, retaining the population in the homeland 3
and the infrastructure is improved as well as all other activities which contribute to prosperity of the region and the country. The ground for enhancing all that lies in the quantity of tourists and travelers. Yet, attracting a bigger number of tourists is not a trouble-free process, particularly in times of ever-changing travel preferences. The rapid development of the Internet, particularly in the past two decades, has changed tourism consumer behavior dramatically (Mills and Law, 2004). It had an enormous impact on tourism industry, specifically to the way how tourists search for information. Moreover, the Internet has influenced tourism in significant manner by providing a great variety of services and products on-line (Kabassi, 2010). So, the Web became the leading source of information particularly important in times of increased number of competitors in tourism market. It was detected as the only way-out to be steady-ready to take prompt action. With the increased importance of search in travelers‟ access to information tourist destinations and businesses were forced to detect more adequate approaches to adapt to the fast-pace change in the environment (Pan et al, 2011). This particularly addresses the on-line tourism supply since tourist destinations have a strong need to acquire data for potential and present tourists and travelers. By the mediation of digital environment, what is noticeable is the obvious tourists‟ transformation from “passive audiences” to “active players” (Prahalad and Ramaswamy, 2000). A noteworthy transformation was made from just passive searching and surfing to creating content, collaborating and connecting. Hence, the development of the Internet empowered the "new" tourists who became knowledgeable and ask exceptional value for their money and time (Buhalis and Law, 2008). In this line, the web-booking systems gain in interest as a direction for detecting differences in the ways that active/passive tourists use the Internet for seeking different kinds of information, booking trips, paying and so forth. Despite the variety of options regarding tourist destination or attraction, tourists are frequently not capable to cope with such a huge volume of choice. Moreover, they need advice about where to go and what to see. In tourism domain recommendations may refer to indicate cities to go to, places to visit, attractions to 4
see, events to participate in, travel plans, road maps, options for hotels, air companies, etc. Such scope of work is very often a robust and needs a facilitating factor. Then a recommendation system is introduced with a main aim to assist tourists in finding a way-out in the e-tourism “chaos”. The main idea of the recommendations is to contribute by facilitating personal selection and prevent tourists and travelers from being overwhelmed by a stream of superfluous data that are unrelated to their interest, location and knowledge of a place. So, it is much easier for tourists to access the information they need thus resulting in shorter lead-time for bookings, making last-minute decisions and generally, tailoring their own packages from a suite of options. So, each country makes efforts and attempts in the line of regional and world promotion as attractive tourist destination. On one hand, tourist destination means temporary location where new travelling experiences may be gained, representing attractiveness of a certain destination (Leiper, 1979). On the other hand, attractiveness may be evaluated in many different ways, such as: from the point of view of emotions, experiences, adventures and satisfaction of tourists (Hu and Ritchie, 1993), with respect to the meaning of tourism attractions and business environment (Enright and Newton, 2004) or, by evaluation of different supporting factors which create tourism supply (Uysal, 1998; Dwyer and Kim, 2003). For instance, initially the concept of tourism competitiveness was related to prices (Dwyer et al, 2000), and later on, econometric models were used for the purpose of ranking (Song and Witt, 2000). Undoubtedly the most comprehensive approach is the one which, beside the competitive advantages, takes into consideration the comparative
advantages
as
significant
factors
which
determine
tourism
competitiveness of a certain destination (Ritchie et al., 2001). There is a variety of definitions and approaches, none being correct or false, but rather helpful in formulating hypothesis for proving different aspects of tourism destination competitiveness (Mazanec et al., 2007). Solution is seen in personalization of information delivery to each traveler together with travel history. Yet, advanced tourist information systems must offer 5
more than just relatively static information about sights and places. The way out is detected in application of recommendation systems as a promising way to differentiate a site from the competitors. So, user-generated content will gain in significance thus enabling development of more accurate recommendation systems. This study intends to present and elaborate necessity of introducing recommenders in tourism, by emphasizing the case of Macedonia. In order to meet this aim and objective, the research is structured in several parts. So, besides the introductory part, Section 2 is rich in findings regarding problems and challenges of tourism. Section 3 presents an overview of different approaches that refer to tourism recommendation systems. The recommended algorithm is in Section 4, while the methodology in terms of design and implementation of recommendation systems is set in Section 5. The main research outcomes applicable to the evidence on Macedonia are noted in Section 6. Section 7 gives an exploratory approach regarding the necessity of developing tourism recommender in Macedonia. Section 8 is the last section which includes conclusions and future research directions. Generally, the contribution of this paper lies in the fact that it enriches the poorly-developed empirical academic work within this scientific area in Macedonia. Additionally, the empirical investigation may alarm the relevant tourism-actors in the country that the time has changed and that the on-line experience has shifted from searching and consuming to creating, connecting and exchanging. Previously passive consumers and web surfers are now generating content, collaborating and commentating. So, this research proposes development of national tourism recommendation system only if being prepared in due time, one may struggle the unexpected challenges.
2. PROBLEMS AND CHALLENGES IN TOURISM This section takes into consideration the past, current and future patterns towards tourism trends. In this respect, this part covers the most profound problems that marked the past two decades. Furthermore, the issue of current status of tourism development is observed in world frames, as well as regional and national 6
boundaries. Finally, this part is supplemented with future challenges that are upstanding towards forthcoming tourism tendencies.
2.1. Past patterns in tourism development: world, regional and national perspective Tourism faced many different events in the first decade of the 21st century. Some of them were driven by the emerging markets and the rapid advances in technology, particularly in digital and social media, but also by the economic environment. From the variety of new challenges, some had a profound effect on the world tourism industry, like the devastating terrorist attack (9/11) in 2001; the combined effect of three significant factors in 2003: the Iraq crisis, the SARS outbreak and a persistently weak global economy; and the global economic recession that started in the second half of 2008. Chart 1 gives a glance of the most significant damaging events in the past two decades. As visually presented the global financial crisis is legitimately identified as the most profound attack on the world tourism development. It can be seen that the growth in receipts closely follows the growth in tourist arrivals with extensive upward trend, particularly in 2009. Chart 1. International tourist arrivals and receipts, World (% change)
Source: UNWTO. (2012a: 5). Note: *projected values 7
Despite the point that tourism in the world has been experiencing continuous growth initiating a positive economic development in the majority of tourismoriented countries, the progressive trend was interrupted by various negative events. Among all, the global financial crisis starting in 2008 and erupting in 2009 had the most negative impacts thus infecting all travel and tourism-related areas. In this respect, the Chart 2 clearly describes the destructive effects which were spread over fifteen consecutive months in terms of international tourist arrivals.
Chart 2. International tourist arrivals, monthly evolution, World (% change)
Source: UNWTO. (2012a: 5) Note: *projected values
The historic shock inflicted by the global financial crisis, has led many countries in the world to unsustainably high levels of public debt, distressed private-sector balance sheets and a surge in unemployment. In this respect governments of different countries took particularly active role in supporting tourism impacts to achieve overall economic development in times of the crisis. Table 1 gives an overview of the state intervention regarding tourism taxes in 2009. Some positive examples can be seen in the cases of Great Britain, Czech Republic, France and Belgium when taxes were reduced for 2-15%. In these cases the governments decided to assist their tourism industries to easier and quicker recovery by decreasing the taxes referring to tourism and hospitality services. However, the bottom rows in Table 1 present some negative examples in terms of tax increasing. Namely due to shocks of the global 8
financial crisis, the governments of Estonia, Lithuania, Latvia and Hungary decided to increase their taxes in order to help national economies in their recovery.
Table 1. Tourism taxes in selected countries, 2009 Country
Tourism taxes
Tourism taxes after
before the crisis
the crisis (%)
(%) Great Britain
17.5
15
Czech Republic
19
9
France
19.5
5.5
Belgium
21
6
Estonia
5
9
Lithuania
5
19
Latvia
5
21
20
25
Hungary
Source: Authors‟ own notes based on www.hotrec.eu Referring to Macedonia the government just recently, in 2010 decreased the VAT rate from 18% to 5%. However, this measure was scheduled and introduced just after the parliamentary elections and despite the positive impacts on tourism development, provoked negative reactions in public being labeled as populist policy. Additionally starting from 2010 the government through the Ministry of economy provides financial support to tourist and travel agencies which promote Macedonia as a tourist destination by incoming tourism. More precisely, subsidies are introduced for bringing organized groups of ten tourists with minimum 3 overnights at categorized accommodation capacities. An alternative is set for the round-tours which have organized tourist arrival and departure by plane, bus or train with minimum 2 overnights in different destinations within Macedonia. Due to the fact that foreign tourists from neighboring countries are dominant, it is normally to have the lowest subsidy of €10 relating tourists from Albania, Bulgaria, Serbia, Greece, Montenegro and Bosnia and Herzegovina. In the line of expending the international tourist 9
market, the subsidies increase to €20 per tourist when coming from Turkey, Romania, Hungary, Slovenia and Croatia. These kinds of measures and activities have long tradition in many countries in line of supporting tourism development. Tourism demand can be measured in variety of units, including national currency, arrivals, nights, days, distance travelled, passenger-seats occupied etc. In order to perceive what was happening in the past decade the number of international tourist arrivals is used as a variable to match the world tendency towards the trend line of Macedonia. As presented in Chart 3, both observed samples have similar trend line representing continuous upward trend. The exception is seen in the World‟s trend in 2003 due to the negative impacts of the Iraq crisis and the SARS outbreak. These events had negative effects on Macedonia as well, but they were postponed for the year to follow so, the number of foreign tourists stagnated in 2004. However, the general conclusion that the financial crisis had superior negative impacts over the world economy starting from 2008 cannot be shared with Macedonia as well. Namely as Chart 2 describes, Macedonia was not faced with intensive negative shocks in the national tourist market in terms of tourist arrivals. Moreover an upward line, but in a slight manner, was noted in 2010 escalating in 2011. Chart 3 International tourist arrivals: World vs. Macedonia, 2002-2011
Source: Authors‟ own calculations based on State Statistical Office. (Various years, various publications) and UNWTO. (2012a: 5). 10
In order to gain more sustainable facts for this first impression regarding the absence of negative influence over the Macedonian tourism, additional analyses were performed. Chart 4 gives a visual presentation of another interesting variable: the average length of stay in the past decade. It is evident that negative trends in the international tourist arrivals are simultaneously followed by decrease in the average length of stay, and vice versa. The most profound negative outcome is noted in 2010 of only 2.14 days average length of stay being almost equal as in 2001 when was calculated to 2.15 days average length of stay. So, the negative effects of the world economic crisis being noted in 2010, have the same destructive outcome as the ethnic war conflict in 2001 in Macedonia. Yet, the average length of stay of foreign tourists in Macedonia for the period 2000-2011 of only 2.22 days is very modest and lower for more than two times, compared to the average length of stay of domestic tourists.
Chart 4. International tourist arrivals and average length of stay in Macedonia, 20002011
Source: Authors‟ own calculations based on State Statistical Office (various years, various publications).
When one wants to analyze the economic importance of tourism what needs to address firstly, is the issue of tourism contribution to the overall economic activity. Therefore, two important economic indicators are analyzed: the gross domestic product (GDP) and the employment, both addressed in tourism. So, Chart 5 presents 11
the GDP and the employees only for hotel and restaurant sector in Macedonia during 2008-2011. More precisely, Chart 5 shows the annual growth of GDP and employees in tourism in Macedonia, which visually supports the statistical glance. Namely the structural breakdown in 2010 is noticeable as a result of the financial recession. One may confirm that the world economic crisis had really intensive negative influence over the tourism industry in Macedonia. This conclusion is additionally supported with another complementary analysis, referring the number of employees in tourism. The average percentage of tourism employment in total labor during the data set is 3.2 %. Although this result might seem moderate, what should be pointed out is that tourism in Macedonia has a higher influence on entire employment in comparison to wider region. Namely, the national average is more than twice bigger than the average of Central and Eastern European countries (CEE) being 1.4% in 2009 (WTTC, 2009b). Yet, the applied official statistical data must be interpreted with a high caution since it does not include unregistered employees in tourism. Moreover, this overview assisted in finding out whether tourism can contribute to job creation thus acting as a factor for decreasing the high unemployment rate of approximately 35%. However, the lack of appropriate statistical data appeared to be a serious obstacle and a crucial limiting factor for more in-depth analysis.
Chart 5. GDP and employees in tourism in Macedonia, 2008-2011 (annual growth %)
Source: Authors‟ own calculations based on State Statistical Office (various years, various publications).
12
Further step in disentangling negative effects and shocks on tourism flows in Macedonia is the analysis of balance of payments for the period 2002-2010. In order to have a clearer general picture for tourism inflows in Macedonia, what should be pointed out is that in 2009 they represented 26% of total inflows of services and 8% of exports of goods. Simultaneously in 2009 the tourism inflows were 20% higher than the foreign direct investments in Macedonia. Within the framework of services, tourism inflows were the second biggest item (just a little bit lower compared to the inflows of transport services), which is 1.3 times higher than the inflows of business services and 2.4 times larger than communication services inflows. When calculated on net-basis, the tourism inflows are far the most important item in the sub-balance of services (Petrevska, 2010). Despite the fact that in the past years the tourism inflows were almost 3 times higher compared to the beginning years of the sample period, yet the importance of tourism in the balance of payments in Macedonia is reduced a lot by tourism outflows. Actually Chart 6 represents that in the early 2000s tourism inflows were almost identical with the outflows. Hence, for some significant net foreign exchange effect of tourism can be discussed only in the middle of 2000s, as a result of more representative inflows of foreign tourists. More precisely as of 2006, tourism inflows in Macedonia gain in importance when they finally exceeded € 100 mil.
Chart 6. Balance of payments - Tourism services in Macedonia, 2002-2010
Source: Authors‟ own calculation based on National Bank of Macedonia, Various publications. 13
Consequently in 2010, they were approximately the same amount as in 2008, meaning that the same level was reached as before the global financial crisis. On the other hand, it is worth mentioning that the average annual net tourism inflows are approximately €39 million, meaning that tourism in Macedonia started to note first significant results. However, it is obvious that in 2009 the inflows were reduced for 30% and the outflows even for 40%, meaning that the foreign and domestic tourists were affected by the crisis. This might lead us to false conclusion about increasing the net tourism flows in Macedonia in times of world recession.
2.2. Current status of tourism development Tourism has emerged as one of the major industries in the world economy by benefiting transportation, accommodation, catering and many other sectors. Thus each country insists on developing it and making a profit from its variety of impacts. Moreover, everyone is interested in increasing the number of incoming visitors since it serves as a source of economic growth. The demand for international tourism maintained momentum in 2011. International tourist arrivals grew by 4.6% to reach 983 million worldwide, up from 940 million in 2010. Europe, which accounts for over half of all international tourist arrivals worldwide, was the fastest-growing region, both in relative terms and absolute terms (29 million more visitors). In 2011 international tourism receipts reached a record €740 bn., up from €699 bn. in 2010. This represents a 3.9% growth in real terms (adjusted for exchange rate fluctuations and inflation), while international tourist arrivals increased by 4.6% in 2011 to 983 million (UNWTO, 2012b: 5). Furthermore in 2011, tourism contributed almost €4.6 trillion to the world global economy, or 9% of the GDP, 100 million direct jobs and €500 billion investments in tourism (WTTC, 2011a: 2). Generally, tourism‟s contribution to worldwide GDP is estimated at some 5%, while contribution to employment tends to be slightly higher and is estimated to 6-7% of the overall number of jobs worldwide (direct and indirect). For advanced diversified economies, the contribution of tourism to GDP 14
ranges from approximately 2% for countries where tourism is a comparatively small sector, to over 10% for countries where tourism is an important pillar of the economy. For small islands and developing countries the weight of tourism can be even larger, accounting for up to 25% in some destinations. (UNWTO, 2012b: 3). It is more than obvious that tourism has experienced continuous expansion and diversification becoming one of the largest and fastest growing industries in the world. Despite occasional shocks, tourism development noted virtually uninterrupted growth. With regards to Macedonia, after a moderate output decline of 0.9% in 2009, the recovery in 2010 was weaker than expected, with the GDP increasing by only 0.7% instead of the expected 1.3%. Inflation accelerated markedly during the year, accelerating from close to zero percent at the beginning of the year to 3.7% in December, leading to an annual average inflation rate of 1.6% in 2010, compared to 0.8% in 2011. Overall the average annual inflation accelerated, from -0.8% in 2009 to 1.6% in 2010. (European Commission, 2011). However, the recovery of negative turbulent shocks from the world financial crisis was noted in the first months of 2011. Consequently, positive upwards trends were noted in terms of tourist arrivals, overnights, average length of stay, GDP as well as the number of employees in tourism sector (more details are already discussed in previous section). Being identified as one of the most promising industries that mainly contribute to the world‟s economy, tourism has become a challenge for every country. Small and developing countries are particularly interested in taking advantages of all positive impacts that tourism implies. In this respect, Macedonia identified tourism as a mean for generating various micro and macro-economic impacts. Consequently, a National Strategy for Tourism Development 2009-2013 was prepared with a main vision - Macedonia to become famous travel and tourist destination in Europe based on cultural and natural heritage (Government of the Republic of Macedonia, 2009: 3). Up-to-date, tourism in Macedonia has accomplished an average growth of 4.64% per year, which is higher than the average growth of the entire economy (3.12%). One may say that the contribution of tourism in GDP is very modest with an average of 15
only 1.7 % per year, but the impression is completely opposite when compared to the average for CEE of 1.6% (WTTC, 2009: 6). Regarding the participation of tourism employees in the total workforce of Macedonia, the national average is 3.1%, which is more than twice bigger than the average of CEE being 1.4% in 2009 (WTTC, 2009: 6). Additionally, the importance of tourism for the national economy can be evaluated by tourism inflows which in 2009 represented 26% of total inflows of services and 8% of exports of goods in Macedonia. In the same line, tourism inflows were 20% higher than foreign direct investments. Accordingly the net tourism inflows in Macedonia have an average of 1% of GDP (Petrevska, 2010).
2.3. Future challenges In order to reduce the risks of making decisions for the future, the forecast of tourism demand is helpful as a foundation on which all tourism-related business decisions ultimately rest (Song and Turner, 2006). Moreover, the forecasts are applied to predict the economic, social, cultural and environmental consequences of tourists and travelers (Frechtling, 2001). Yet, there are varieties of changes in the surrounding which often cannot be envisaged, like financial crises, terrorist attacks, war conflicts, epidemics etc. Even when an ideal forecasting model is identified, it can only serve as an approximation for complex tourists‟ behavior, for it is possible that tourists‟ decisions change reflecting the changes in preferences, motivation or economic shocks (Hall, 2005).
Chart 7. World GDP forecast by industry, 2012-2022
Source: WTTC. (2012: 4). 16
Chart 7 gives a look on some latest projections towards tourism contribution to the GDP. Namely, it is estimated that tourism influence on GDP to grow at a rate of 4.2% on annual basis over the next decade being faster than automotive, education and mining sector as well as the total economy which are expected to grow by 3.6% per year (WTTC, 2012: 4).
Chart 8. Outlook: World
Source: UNWTO. (2012a: 6). Note: *projected values
The updated analyses of the world leading tourism experts confirm confidence weakening, but still with positive patterns. In this respect, the projection for the full year of 2012 for the international tourism in the world is foreseen to decline from 4.4% in 2011 (WTTC, 2011b) to 3-4% in 2012 (Chart 8). In absolute figures, it is expected that in 2012, 1 billion tourists will be involved in travel and tourism activities around the world. Furthermore in these frames, the Europe is forecasted to mark the sharpest decline from 6% in 2011 to 2-4% in 2012. However, most worryingly, the last world tourism leading panel did not propose anything to address the main short-term risk, pointing to the danger of a sovereign funding crunch in the early 2012. The potential crisis may spill over to the real economy as banks tighten the credit standards and the business confidence weakens. This is particularly referred to the Eurozone economy, which has potentially gone 17
back into recession again and has a stagnation forecasts for 2012. This kind of economic backdrop is incredibly challenging environment for tourism. When addressing the challenges in a long-term perspective, a positive upward line is forecasted. Namely, based on a linear trend, it is forecasted that the world‟s international tourist arrivals from 1 bn. in 2012, will reach its highest point of 1.8 bn. in 2030 (UNWTO, 2012b: 13). The long-term quantitative forecasts are visually presented in Chart 9.
Chart 9. World actual trend and forecast, 1950-2030
Source: UNWTO. (2012b: 14).
With regards to the case of Macedonia, although many negative effects on tourism development were previously mentioned and elaborated, the forecasts are much more optimistic than the actual outcomes. Namely, the estimated results are encouraging and by 2021 it is expected that the direct contribution of tourism to GDP will have reached 1.6 %, thus bringing revenue of €170 million according to constant 2011 prices; the total contribution of tourism to GDP will have risen to 6.0%; the visitor exports are expected to generate €75 mil. (5.1% of total exports); and the investment in tourism is projected to reach the level of €76 million representing 2.8% of total investment. Additionally, it is expected that the number of employees that 18
indirectly support tourism industry in Macedonia will have an upward trend and will reach 35000 jobs in 2021, representing 5.4% of total workforce (WTTC, 2011a). Speaking about the international tourism demand, the upward trend is expected to continue in the next period (Petrevska, 2012a). So, when applying the same forecasting method, as in the long-term projection for the world‟s international tourist arrivals by 2030, we can expect an increase of almost 2.5 times. In this line, by modeling tourism demand with a linear trend line, an upward trend is projected up to 634574 arrivals (Petrevska, 2012b). Likewise, it should be pointed out that the anticipated values must be taken into consideration with a large doze of precaution, because they do not indicate the reasons which affect the forecasted results. Beyond that, these findings underline the fragile nature of tourism industry and its affection from strong negative events like the world financial crisis. This optimistic view is supplemented additionally with the fact that the number of user ratings is permanently increasing by 15% monthly growth rate. Supportive but not surprising, is another fact noting an upward trend of web portal users, which complements the positive general conclusion referring tourism income in Macedonia. The average tourism consumption of €50 per day (WTTC, 2010) is anticipated to note an increase of one third of a euro, which may be misinterpreted as insignificant to national economy. However, on a long-term horizon based on these projections, the tourism contribution to GDP may note an increase of more than 1%. Consequently, Macedonia identified tourism as an industry which might contribute to enhance the foreign export demand for domestic goods and services, generating foreign currency earnings, new employment opportunities, repaying the foreign debt, increasing the national income etc.
2.4. ICT and e-tourism This part starts with brief overview on the subject: information and communication technology (ICT) vs. e-tourism. In this line, it may be noted that in short time, the Internet was introduced as a rapidly evolving medium for travel and tourism (Schonland and Williams, 1996). Its successful introduction to e-tourism is 19
fully supported by the search engines which became a dominant source in tourists‟ use to access particular tourism and travel products. Due to its significance, this issue raised an interest among academia and practitioners. Generally, they argue regarding the understanding how search engines work and how travelers use the Internet and booking systems as tools in e-tourism (Morrison et al., 2001; Singh and Kasavana, 2005; Connolly and Lee, 2006; Pan et al., 2007; Buhalis and Law, 2008; Pan et al., 2011; Xiang and Pan, 2010). Moreover, the success of the search engine marketing requires good understanding of consumer behavior in order to provide the information desired by different consumers. Furthermore, the necessity for developing digital technology that will support the personalized services to address individual needs is fully justified. Tourism actors should collect customer information before, during and after a visit, in order to better understand consumer behavior choices and determinants (Buhalis and O'Connor, 2005). Additional insights regarding the progress of IT in tourism domain is noted in many research findings (Kluge, 1996; Kirk and Pine, 1998; Frew, 2000; O‟Connor and Murphy 2004; Leung and Law, 2005 and 2007; Law et al., 2009). Some researches address different approaches dealing with variety of relationships that appeared in e-tourism. So, Weber and Roehl (1999) explored demographics between Internet users and tourists at the same time. However, little research has been done on the travel-related behaviors of Internet travelers. In this respect, Morrison et al. (2001) found that some travelers book on-line, while others go to travel agents or call the toll-free numbers of travel providers after getting travel information on-line. With regards to the behavioral dimensions, it may be utilized to segment travel markets as a powerful tool in managing e-tourism (Hennessey et al., 2008). Regardless the approach, it must be underlined that tourism needed this kind of information some years ago, while today we are faced with tourists with different travel patterns provoking different activity while travelling.
20
3. EXAMPLES IN TOURISM RECOMMENDATION SYSTEMS This section first gives a short introduction to recommendation systems, some basic concepts and definitions related to this kind of system. In the following, it presents an overview of different approaches that refer to tourism recommendation systems. More precisely, this part encompasses some interesting and meaningful aspects that tangle the issue towards the necessity of introducing recommenders in tourism domain. In this respect, this section begins with introductory notes regarding the recommenders, basic terms and concepts, as well as the recommendation process. Furthermore, this part gives a glance on necessity of identifying tourists‟ and travelers‟ preferences, as a precondition in creation of specific, tailor-made tourist products. Finally, this segment encompasses tourism recommenders, as intelligent applications that offer responses to tourists‟ and travelers‟ queries.
3.1. Introduction to recommendation systems Nowadays, there are a lot of recommendation systems accessible via internet, which attempt to recommend the users several products such as music, movies, books, etc. In order to understand them, first, it is necessary to have a description. In a general way, recommendation systems are systems which intend to acquire opinions or preferences about items from a community of users, and use those opinions to present other users with items that are interesting to them. From this general description we can see that recommendation systems need two basic things to work properly: information about the preferences of the users and a way to determine if an item is interesting for a user. Normally, the users‟ information includes external information, such as user profiles, purchases histories, and product ratings (Orio, 2006). The way to determine, whether an item is interesting to a user or not, depends on the kind of recommendation system and the techniques used to find similarities among items or users. This description is quite general and could be applied even to persons that recommend items to other persons. A more specific definition of recommendation systems is given in (Setten, 2005). 21
“Systems that produce individualized recommendations as output or have the effect of guiding the user in a personalized way to interesting or useful objects in a large space of possible options.” The main keywords in this more formal definition are individualized and personalized. These terms indicate that every user will be presented with different information sources or items, depending on the information the system has about every user. In order to continue the discussion about recommendation systems, we will have to define several terms that will be used through the rest of this document.
3.2. Terms and concepts The following terms are often used in a recommendation system and the definitions introduced here are based on the work presented in (Setten, 2005). Item: in the context of recommendation systems, an item represents the information the system possesses about any object. An object can be an electronic document, a product, a person, a service or anything that can be represented by information. Recommender: a recommender is any entity that gives personalized recommendations as output to users‟ preferences. It may be possible that a recommender does not produce a specific output, such as a list, but they might guide somehow the users in an individual way to useful or interesting items. A recommender could be a person or a software system. Recommendation: this is the output of a recommender; it can be compound by an item or a list of items. The items presented to the users have to be interesting to them, according to the recommender. The criterion used to determine if an item is interesting or not for a user, depends exclusively on the technique used by the recommender. User‟s Interest: this is an abstract representation of how much a user appreciates an item. This is a subjective concept and it is hard to represent it in an objective way. 22
Prediction: the expected interest of a user in one item. This concept is different from the concept of recommendation. While some systems might present predictions
with
the
actual
recommendations,
others
can
produce
recommendations only. Rating: an objective measure representing a user‟s interest. The possible values of this measure are given according to the scale established by the designer of the recommendation system. Predicted Rating: an objective measure representing the expected interest of a user in certain item. This measure is estimated by the system and its possible values are elements of a specific scale. Actual Rating: objective measure representing the real interest of the user in a specific item. This value is given by the user himself according to the scale of rates of the system. Prediction Accuracy: a measure that indicates at which extent the predicted rating agrees with the user‟s actual rating. The more accurate the predictions the better the performance of the recommendation system. Prediction Technique: the specific algorithm that the recommendation system will use in order to calculate the predicted rating of an item.
3.3. Recommendation process Once we have defined some concepts and terms generally used in recommendation systems, it is necessary to explain the way these systems work and produce recommendations to users. In a general way every recommendation system follows a specific process in order to create recommendations. Further follows an explanation of this process and steps. If we see the process of recommendation as a black box, as shown in Figure 1, we can identify two sources of information needed as input for the process. These sources of information are the users‟ profiles and the information about items or products. Ideally, the information stored in the profiles is related with the preferences of the users and should be given explicitly by the user himself. However, this 23
information can also be extracted from other external sources, such as web pages, buying behavior, etc.
Figure 1. Recommendation process as a black box
The information about the items can range from special metadata of the product, information extracted from the item or the item itself in the case of electronic documents. In the case of tourism recommenders, this information may produce databases of huge dimensions. In Figure 1 we can also distinguish that the final product of the system will be a set of recommendations for the user. The final representation of these recommendations depends on the system itself, but it may range from ordered lists of items, brief description of the items, or the items represented on the map. The recommendation process in a more detailed way is shown in Figure 2. It includes the following steps: information recollection, selection, transformation, structuring and presentation. From all the steps presented in Figure 2, the information recollection step is the only one that is not done by the system itself. Figure 2 gives a small description of each of these steps.
24
Figure 2. Recommendation Process
3.3.1. Information recollection Although the information recollection step is not performed by the recommendation system itself, it is a really important step. It includes the recollection of users‟ personal preferences and information about items such as metadata, opinions, features extracted directly, etc. This step has a special importance since it will be the base for the whole recommendation system. If the information collected presents incongruence or contradiction, the system will not be able to produce not even regular recommendations. For this reason, special attention should be put on collecting information that truly reflects the preferences of the users, or information that truly represents the items.
3.3.2. Selection According to (Setten, 2005) the step of selection consists of determining “which items are interesting or relevant enough for a user and removes all other items from the retrieved set of items”. The way the selection of the items is done depends strictly on the approach taken to find items that are similar to the ones the user considers relevant. Therefore, the key concept in this step is how we define similarity between items. The concept of similarity could be defined in terms of other users‟ profiles, of features extracted from the items, or in terms of metadata associated with the items. 25
We will not define the concept of similarity yet, since it depends on the approach taken.
3.3.3. Transformation The main objective of the transformation step is to perform some modifications to the items retrieved. This step is only optional in the general recommendation process and it addresses transformations such as summary, change in the quality of the items, creation of details for its further presentation, etc.
3.3.4. Structuring The step of structuring is related to the construction and organization of the structure that the user will use to navigate through different recommended items. This step can include activities, such as grouping the items according to certain characteristics, sorting the groups of items, sorting items inside these groups, linking items that have some relationship, etc.
3.3.5. Presentation The final step in the recommendation process is associated with the presentation of the different retrieved and structured items to the final user. According to (Setten, 2005) it “deals with issues such as layouts, document formats, colors, fonts, and presentation medium”. This final step in the process should be designed and executed carefully, since the user will interact with the system by means of its results.
3.3.6. Feedback One additional step, which might be interesting to consider, is the one related with the feedback to the system. Although this step is optional, it can help greatly to improve the results of the recommendation system. With an appropriate feedback any of the previous steps could be improved. The type of feedback obtained from the user could be of two kinds: implicit or explicit. In explicit feedback, the user provides the system information about how relevant the recommended items are. On the other 26
hand, implicit feedback is obtained from the user by analyzing his usage behavior, for instance, how much time he spends on looking at the retrieved items. 3.4. Tourists’ preferences and related work It is more than obvious that whether a potential tourist will be interested in a certain item depends on the preferences. It may sound fragile, but the vast majority of today‟s tourists know exactly what they are looking for. Yet, they are very demanding and have complex, multi-layered desires and needs. Nowadays, the so called “postmodern tourists” have specific interests and individual motifs which result in tailor- made tourist products according to their particular preferences. They are often high experienced in travelling and demand perfect tourism products rather than standardized ones. Consequently, they take much more active role in producing diversified tourism products with shorter life cycles, which is enabled by increased usage of the information technology. Many researchers were interested in identifying tourists‟ needs, expectations and behavior. In this respect, numerous papers discuss tourist roles in order to define their considerable variations. In mostly, the behavior is related to specific demographic and background characteristics emphasizing the life course as the leading component for investigating tourist role preferences. Yet, attention should be paid to a variety of social structures and processes, including psychological needs and life-course stage. Cohen (1972) was one of the first sociologists who proposed a typology to conceptually clarify the term “tourist” by developing a four-fold typology. Based on that, Pearce (1982) identified specific behaviors thus enabling tying the evolutionary nature of tourist role preference and the psychological needs. Moreover, he developed fifteen different tourist types which allowed creation of several measurement scales. In this respect, the Tourist Roles Preference Scale (Yiannakis and Gibson, 1992) presents a comprehensive classification of leisure tourists. Additional work resulted in adding two more tourist types to the tourist categorization (Gibson and Yiannakis, 2002). Moreover, researchers focused on exploring the experience of tourists as well as the importance of the tourist experience for tourists (Yfantidou et al., 2008). 27
On the other hand, some researchers emphasize the relationship between ICT and tourists‟ behavior. Namely, many of them underline that ICT acts as a protector and enhancer, thus directly having influence on tourists experiences, preferences and behavior (Winata and Mia, 2005; Singh et al., 2006; Kim and Ham, 2007). Furthermore, some researchers emphasize the possibility of personalizing tourism customers‟ experience by application of ICT (Niininen et al., 2007).
3.5. Tourism recommendation systems and related work One may argue the inevitable relationship between tourism and information. Moreover, it is widely‐recognized fact that information and decision‐making have become the foundation for the world economy (Wang, 2008). Due to the importance of tourism, recommendation systems applied in tourism have been a field of study since the very beginnings of artificial intelligence. It is a matter of identifying a class of intelligent applications that offer recommendations to travelers, generally as a response to their queries. They mostly leverage in-built logical reasoning capability or algorithmic computational schemes to deliver their recommendation functionality. So, the recommenders attempt to mathematically model and reproduce the process of recommendations in the real world. Due to the rapid expansion of e-tourism, the tourism recommendation systems attracted a lot of interest in academia. In this respect, Mirzadeh et al., (2004), McSherry (2005) and Jannach (2006) elaborate the need for developing intelligent recommendation systems which can provide a list of items that fulfill as many requirements as possible. Ricci et al. (2002) and Wallace et al. (2003) discuss a recommender system dealing with a case-based reasoning in order to help the tourist in defining a travel plan. However, the most promising recommendation systems in the tourism domain are the knowledge-based and conversational approaches (Ricci and Werthner, 2002; Thomson et al., 2004). Yet, some other variants of the contentbased filtering and collaborative filtering are engaged for recommendation, like knowledge-filtering, constraint-based and case-based approaches (Kazienko and Kolodziejski, 2006; Ricci and Del Missier, 2004; Zanker et al., 2008). In the same 28
line, the recommendation systems based on a text mining techniques between a travel agent and a customer through a private Web chat may easily be applied (Loh et al., 2004). Some recent academia work refers to more sophisticated outcomes than the above noted. Namely, the introduction of a personalized tourist information provider as a combination of an event-based system and a location-based service applied to a mobile environment is suggested by Hinze et al. (2009). On the other hand, an investigation on sources and formats of on-line travel reviews and recommendations as a third-party opinion in assisting travelers in their decision making during the trip planning is presented in the work of Zhang et al. (2009). Noticeable are the findings regarding development of a web site in order to enable Internet users to locate their own preferred travel destinations according to their landscape preferences (Goossen et al., 2009). Furthermore, the usage of the orienteering problem and its extensions to model the tourist trip planning problem is elaborated as efficient solution for number of practical planning problems (Vansteenwegen and Wouter, 2011). It is evident that the research area is extending and resulting in improving dependability of recommendations by certain semantic representation of social attributes of destinations (Daramola et al., 2010). Moreover, most of the recommendation systems focus on selecting the destination from a few exceptions (Niaraki and Kim, 2009; Charou et al., 2010).
4. RECOMMENDATION ALGORITHM The main objective of our research is to develop a tourism web portal which relies on an efficient and accurate personalized recommendation algorithm that will support tourists visiting certain state or region by helping them to identify relevant tourist objects matching their personal interests and planning their trips more efficiently. To accomplish this objective, a several step algorithm was developed. This chapter aims at describing the developed recommendation algorithm used in our system. 29
Table 2. Tourist types as defined by (Gibson and Yiannakis, 2002) No. Tourist type Sun Lover 1 2 3 4 5 6 7 8 9 10 11
12 13 14 15 16 17
Description Someone primarily interested in warm places with lots of sun, sand, and ocean. Action Seeker Interests include partying, going to night clubs, and having uncomplicated romantic experiences. Anthropologist Interests include meeting local people, trying their food, and speaking the language. Archeologist Interests include archeological sites and ruins; enjoys learning the history of ancient civilizations. Organized Interested in organized vacations, package tours, taking Mass Tourist pictures, and buying lots of souvenirs. Thrill Seeker Interested in high risk, exciting activities that bring them to emotional heights such as base jumping. Explorer Enjoys adventures, going out of the way places and the challenges involved in getting to those places. Jet Setter Enjoys going to high class hotels and resorts as well as exclusive night clubs and talking to celebrities. Seeker Interested in seeking knowledge to better understand one‟s self and or the meaning of life. Independent Likes to visit regular tourist attractions but plans own Mass Tourist trip with minimal assistance from others. Independent Likes to visit regular tourist attractions, does so in a Mass Tourist highly spontaneous manner with very little planning. II High Class Appreciates first class travel, hotels, and restaurants, as Tourist well as seeing shows. Drifter Enjoys nomadic experiences; lives a hippie life style. Escapist I Interested in vacationing to get away from the daily routine and the stresses of everyday life. Escapist II Interested in places of solitude which are quiet and peaceful Sports Lover Primarily engages in very active vacations where they can still play their favorite sports. Educational Participates in planned study tours and seminars to Tourist obtain new knowledge and skills
The first step foresees tourist and tourist objects profiling. Tourist profiling is a two-step process which involves creating the profile and then reviewing the profile to make any necessary adjustments. The system uses tourist types taken from the scientific tourism literature to model the tourist personal profile. The tourist profile is 30
defined as an N-dimensional vector. Vectors are suited to model such tourist profile whereby each dimension corresponds to a certain tourist type while the value indicates how much tourist identifies him- or her-self with corresponding type i.e. indicates the degree to which tourists identify themselves with the given types as defined by (Gibson and Yiannakis, 2002). They measured “tourist role preference, satisfaction of psychological needs, and demographic characteristics such as age, gender, marital status, and education level”. The conclusions of their study resulted in creating 17 tourist types as explained in Table 2. Typically, individual tourist cannot be characterized by only one of these archetypes but, has unique combination of these personalities though to varying degrees. Thus, tourist types model the tourists‟ generic interests in an abstract form. The process of user profiling using vectors is presented in Figure 3. Figure 3 depicts an exemplary tourist, who likes to enact in the role of an adventurer, followed by sport and cultural, and rather dislikes sightseeing activities.
Figure 3. Modeling the tourist profile using vectors
The initial tourist profile for each system user is created by the user himself during the process of registration, by determining the degree of membership to each of the tourist types. Considering the fact that the human preferences change over the time due to various factors, the tourists might change their behavior too. To make the 31
system capable to cope with these changes, we have enabled tourist profile adjustment. It is based on the ratings the tourist gives for each tourist object that he visits after his journey and according to formula (1).
Uijt 1
1 (Uijt Rikt 1 * w * Okj) 2
(1)
Where Ui represents the i-th user and Ui U , U- is the set of users registered to the system, Uijt is the degree of membership in the moment t of the i-th user to the tourist type Tj and
Tj T ,
T – is the set of tourist types according to literature (Gibson &
Yiannakis, 2002). Ok O represents the k-th object in the set of all objects O registered in the system, w-is the weighting factor and Rik is the rating of the k-th tourist object given by i-th user. Similarly, we may generate profiles for attractions and in the same way as the tourist profile is represented in form of a vector; every tourist object is modeled through a vector as well. Thereby, in a quantitative way this vector describes how much the object is related to the given types. For example, the famous monastery Saint Panteleimon in the city of Ohrid known as a birthplace of Cyrillic alphabet and used by St. Kliment for teaching the Cyrillic alphabet, might be highly relevant for sightseeing tourists but not for such kind of tourists that would like to do some risky activities. In the developed system a manual process to link the given tourist types to appropriate tourist objects is proposed. Therefore, for each of the tourist objects, the degree of relationship to each of the tourist types is specified by domain experts. In order to prevent information overload of the tourist and provide only relevant information, the system should recommend a subset of tourist objects according to the personal experiences, individual tourist desires and those he/she prefers to avoid. This in turn might lead to an increase of the tourist's satisfaction of experiencing a relaxed sightseeing trip.
32
According to this, the next step of the proposed methodology aims at matching tourist profiles against the set of tourist objects on the basis of tourist types, thus producing a ranking list of objects for each given tourist and reducing the set of objects. If a tourist profile matches the characteristics of an object, this object should be recommended to the respective tourist. Therefore, the matchmaking algorithm has to examine whether they share similar structures. The more similarities they have in common, the more contributes the tourist object to the tourist‟s satisfaction and therefore should be ranked higher. To estimate the similarity degree between tourist profiles and tourist objects, the system contains a special module based on a vector-based matchmaking function, whereby a given profile and each tourist object constitute vectors and are compared in a vector space model. A common method to obtain the similarity is to measure the cosine angle between two vectors. If the vector space is non-orthogonal, kernel based algorithms can be applied to measure the similarity in such space. The dimensions of the vector space model correspond to the selected tourists types found in the scientific tourism literature (Gibson and Yiannakis, 2002), such that each distinct tourist type (e.g., adventure or cultural type) represents one dimension in that space. The implemented matchmaking function has the following form (2): N
SIM cos (Ui, Oj )
Ui
k
Ojk
k 1
N
Ui
k
k 1
N
2
Oj
2 k
k 1
(2)
Where Uik is the degree of membership of the i-th user to the tourist type Tk, Ojk is the degree of membership of the j-th tourist object to the tourist type Tk, and N is the number of tourist types. According to equation (2), the degree of similarity between tourist profiles and tourist objects will be calculated. Tourist objects will be ordered by the value of the matchmaking function for a given user, and only those objects that have positive value for this function will be considered for recommendation (3): 33
Oirec Oj, where SIMcos (Ui, Oj) 0
(3)
Considering the five point Likert scale for rating the objects, to each object in the constructed set, a recommendation mark will be assigned, according to formula (4):
Rirec R(Oj) 5 * SIMcos (Ui, Oj), Oj Oirec
(4)
In our methodology, we have considered another very important fact related to the behavior of the people planning a vacation or trip. In everyday life, while planning a vacation or trip, people also rely on recommendations from reference letters, news reports, general surveys, travel guides, and so forth. In addition, they desire personal advice from other people with similar preferences or people they trust. In fact, over 80% of travelers participating in a TripAdvisor.com survey agree that “reading other travelers‟ online reviews increases confidence in decisions, makes it easier to imagine what a place would be like, helps reduce risk/uncertainty, makes it easier to reach decisions, and helps with planning pleasure trips more efficiently” (Gretzel, 2007). Experimental findings show that there exists a significant correlation between the trust expressed by the users and their similarity based on the recommendations they made in the system; the more similar two people are, the greater the trust between them (Ziegler and Golbeck, 2006). Similarity can be interpreted in several ways such as similarity in interests or ratings or opinions. Different methodologies can be used to calculate the similarity between the users in the system. As one of the most prevailing and efficient techniques to build recommender systems, collaborative filtering (CF) implements the idea for automating the process of “word-of-mouth” by which people recommend items to one another. It uses the known preferences of a group of users who have shown similar behavior in the past to make recommendations of the unknown preferences for other users. CF is facing many challenges, among which the ability to deal with highly sparse data and to scale 34
the increasing numbers of users and items as the most important in order to make satisfactory recommendations in a short time period. Sparseness of ratings data is the major reason causing poor recommendation quality. The sparseness problem occurs when available ratings data is rare and insufficient for identifying the similar neighbors. This problem is often very significant when the system is in its early stages. On the other hand, when numbers of existing users and items grow tremendously, traditional CF algorithms will suffer serious scalability problems, with computational resources grown nonlinearly and going beyond practical or acceptable levels.
To reduce the dimensionality of data and avoid the strict matching of
attributes in similarity computation the cloud-model CF approach has been adopted. It is constructing the user‟s global preference based on his perceptions, opinions and tastes, which are subjective, imprecise, and vague (Palanivel and Siavkumar, 2010), and it seems to be an appropriate paradigm to handle the uncertainty and fuzziness on user preference. The main goal of the cloud model CF is to construct the global preference for
each user by calculating a triple of three digital
characteristics V = (Ex, En, He) .
The
expected value Ex represents the typical value of user ratings, that is, the average of user ratings. The entropy En represents the uncertainty distribution of user preference, which is measured by the deviation degree from the average rating. The hyper-entropy He is a measure of the uncertainty of the entropy En, which is measured by the deviation degree from the normal distribution. Given a set of ratings data for a user ui, rui = ru = (ru,1 , ru,2,..., ru,n), the three characteristics can be defined as in formula (5) (Zhang et al., 2007).
1 Ex n En
n
ru ,i
i 1
2
1 n
n
ru ,i Ex
i 1
1 2 1 n He S En , where S (ru ,i Ex) 2 3 n 1 i 1 2
35
(5)
The k similar (neighbor) users, for an active user are selected based on the cloud model similarities between the active user and the users that already rated the object Oj Oirec .
A likeness similarity method based on cloud model using the cosine
measure was proposed in Zhang et al., 2007. Given two cloud models in terms of the
characteristic vectors
V u = (Exu , En u , He u )
and
V v = (Ex v , En v , He v ) ,
the similarity between
them are defined as (6):
Exu Exv Enu Env Heu Hev
sim (u, v) cos(Vu ,Vv )
Exu Enu Heu 2
2
2
Exv Env Hev 2
2
2
(6)
Recommendation function based on the cloud model is defined as (7):
(r
v, j
Ru , j r u
vN ( u )
rv ) sim (u , v)
sim(u, v)
vN ( u )
(7)
Where N(u) is the k most similar users to active user u and ru and rv are the average rating of user u and v, respectively. The value of rating rv,j is weighted by the similarity of user v to user u; the more similar the two users are, the more weight rv,j will have in the computation of the recommendation function. Total recommendation function for a given tourist object (Oj), is calculated using a weighted average of the functions given by equations (2) and (7), thus comprising (8):
Freci , j
w1 * SIMcos (ui , Oj ) w2 * Rui , j w1 w2
(8)
36
According to the value of total recommendation functions the objects will be ordered and further classified into five categories in the following way (9):
Cati , j
k 1, Oj Oirec 0 Freci , j 0.2 k 2, Oj Oi 0.2 Frec 0.4 rec i, j k 3 , Oj Oi 0 . 4 Frec 0 . 6 rec i, j k 4, Oj Oi 0.6 Frec 0.8 rec i, j k 5, Oj Oirec 0.8 Freci , j 1
(9)
5. DESIGN AND IMPLEMENTATION OF RECOMMENDATION SYSTEM This section first presents the system architecture. In the following the main characteristics of the web based GUI are described. Actually, some of the features of the developed GUI are presented in various use-cases. Considering the importance of the system implementation several structures and code patterns are presented at the end of this section.
5.1. System architecture The system architecture is presented in Figure 4.
Figure 4. System architecture
The developed national tourism web portal is lying or is built using the social network framework. Our portal is a significant improvement on existing travel 37
websites and provides tourists a customized, unique, and enriching travel experience. It incorporates some standard plugins typical for social networks like Facebook. But, it advances the concept by including custom plugins, like the recommended objects plugin which is the core of the portal. It is using the Google Map of Macedonia, to visualize both: static tourist objects (object that are not temporary, like churches, museums, archeology localities, etc.) and dynamic object (object that have limited time duration, like events, expositions, etc.).
5.2. Web based graphical user interface The graphical user interface is changing its appearance in accordance with the current users‟ actions. Besides the basic social network plugins and applications, the system implements the recommendation features through two main applications. One of them is dedicated to the recommendation of tourist objects and the second one is dedicated to the trip planning. Both of them are using Google maps API and they display the tourist objects on the map according to their geographical location. The objects are also grouped into contextual layers (cultural, sport, etc.) thus augmenting the Google map layers. Moreover, the objects are also geographically grouped into municipalities.
Figure 5. Recommended municipalities
38
Municipalities are recommended to the user in the form of circles displayed on the map (Figure 5). The size of the circle indicates the user‟s affinity for the municipality; therefore, a large circle indicates a municipality with many tourist objects with high recommendation marks i.e. that match the user profile. By displaying the user‟s affinity through the size dimension of the circle, users can easily observe which municipalities would be of most interest to them. Inside each of the suggested circles most appropriate tourist objects for a given user are indicated with points inside the circles (Figure 6).
Figure 6. Recommended municipalities and tourist objects
When the map is zoomed the tourist objects are displayed as icons in the location of the correspondent object as shown in Figure 7. The image of the icon indicates the type of tourist objects such as a museum, church, or restaurant. The size indicates how closely the object meets the user‟s interests. Each attraction also has an information window as displayed in Figure 7. The information window usually includes the name and picture of the attraction, an icon of an umbrella indicating that the attraction is accessible in the rain, and tags.
39
Figure 7. Recommended tourist objects The information window also displays a general idea of the time consumption of the attraction, friends who have visited the attraction, and an option to view narratives in video, audio, or text format. Through this window, the user can also rate the object. This operation is recommended to be done after visiting the object and according to the personal experience and satisfaction. The goal of this operation is two-fold: to help updating the user profile, and to make the process of recommendation more accurate. Moreover, if the user is planning a trip and enters the preferred period of stay as well as a starting point of his trip, the system will try to suggest him the optimal route to visit as many as possible objects of interest within the limited time period (Figure 8).
Figure 8. Planning a trip 40
The system is taking into consideration the weather conditions, average time necessary to be spent at each object, availability of transportation means and times and dates of the dynamic objects. The route is visually marked on the map. If the tourist is using the application from a smart phone, according to his current position coordinates the closes objects of interest are visualized.
5.3. Implementation issues One of the most important aspects of the implementation of the system is the kind of structure to store the large amounts of data (users, items, ratings and similarities). The main criterion for the choice was to save the largest possible computation time. Because of this, the main data structures are implemented with the class of the .NET framework Dictionary (TKey, TValue). The Dictionary (TKey, TValue) generic class provides a mapping from a set of keys to a set of values. Each addition to the dictionary consists of a value and its associated key. Retrieving a value by using its key is very fast, close to O(1), because the Dictionary (TKey, TValue) class is implemented as a hash table structure. Another advantage of the class Dictionary (TKey, TValue) is its enumeration properties. It is possible to access to all the stored pairs of key-value with the instruction foreach. For the implementation of the User class, several attributes from various types were used. Some of them are presented with Code Listing 1.
Code Listing 1. The User class. 41
The attribute ratingAverage stores the average of the ratings of the user. It is possible to calculate this value at any moment from the list of ratings. As a consequence, an extra computational time is required. The value only changes when the user rates a new tourist object. According to that, the final decision has been made to store this value in an attribute and to recalculate it each time the user modifies his or her ratings. The attribute ratings is implemented with a Dictionary where the Key is the id of the item and the Value is the value of the rating. It gives instant access to the rating of the user for a specific item. The most relevant details of implementation about the methods of the class User are presented in Code Listings 2 and 3.
Code Listing 2. The method for recalculating the rating parameters.
Another important method is the one for calculating the rating average (Code Listing 3).
Code Listing 3. The method of the User class which calculates the rating average.
42
The method RatingAverageWithSkippedItems is overloaded. The method obtains the rating average using the value calculated (ratingAverage) as base. Then, it modifies this value using the values of the ratings of the skipped items. The first method uses the list of skippedItems of the own user that executes the method. It is to calculate the own ratingAverage of the user with his or her skippedItems. However, as said before, the similarity formulas compare two users: the active user (u) and a neighbor (n). For the calculation of the similarity, it is necessary the rating average of the user (n) who is compared to, but excluding the calculation of the skipped values of the user (u). The second method calculates the average rating of the owner but skips the items it receives as parameter from another user. The method CommonRatings presented in Code Listing 4, returns a list with the rating that two users (the owner and another received as parameter) have in common. The method does not add the ratings of the skipped items list of the owner even if it is a common item with the other user.
Code Listing 4. The CommonRatings method.
The class Item keeps the data for describing the tourist objects. It contains several attributes and some of them are presented in Code Listing 5.
Code Listing 5. The Items class. 43
The attribute ratingVariance stores the variance of the ratings of the item. This value is possible to calculate at any moment from the list of ratings, but it requires an extra computational time. The attribute ratings is implemented with a Dictionary where the Key is the id of the user and the Value is the value of the rating. It lets instant access to the rating of the user for a specific item.
6. EMPIRICAL EVIDENCE ON MACEDONIA In this section we describe the experiments conducted in order to evaluate the developed recommendation system. For this purpose offline experiments were conducted. They are typically the easiest to conduct, as they require no interaction with real users. An offline experiment is performed by using a pre-collected data set of users choosing or rating items. Using this data set it was tried to simulate the behavior of users that interact with a recommendation system. In doing so, the assumption was that the user behavior when the data was collected would be similar enough to the user behavior when the recommender system is deployed, so that reliable decisions based on the simulation can be made. Offline experiments are attractive because they require no interaction with real users, and thus allow to compare a wide range of candidate algorithms at a low cost. As the goal of the offline evaluation is to filter algorithms, the data used for the offline evaluation should match as closely as possible the data the designer expects the recommender system to face when deployed online. Care must be exercised to ensure that there is no bias in the distributions of users, items and ratings selected. For example, in cases where data from an existing system (perhaps a system without a recommender) is available, the experimenter may be tempted to pre-filter the data by excluding items or users with low counts, in order to reduce the costs of experimentation. In doing so, the experimenter should be mindful that this involves a trade-off, since this introduces a systematic bias in the data. If necessary, randomly sampling 44
users and items may be a preferable method for reducing data; although this can also introduce other biases into the experiment (e.g. this could tend to favor algorithms that work better with more sparse data). Sometimes, known biases in the data can be corrected by techniques such as reweighing data, though such correction is often difficult. In order to evaluate algorithms offline, it is necessary to simulate the online process where the system makes predictions or recommendations, and the user corrects the predictions or uses the recommendations. This is usually done by recording historical user data, and then hiding some of these interactions in order to simulate the knowledge of how a user will rate an item, or which recommendations a user will act upon. There are a number of ways to choose the ratings/selected items to be hidden. Once again, it is preferable this choice to be made in a manner that simulates the target application as closely as possible. In many cases, yet, we are restricted by the computational cost of an evaluation protocol, and must make compromises in order to execute the experiment over large datasets. Ideally, if we have access to time-stamps for user selections, we can simulate what the systems predictions would have been or had it been running at the time the data set was collected. We can begin with no available prior data for computing predictions, and step through user selections in temporal order, attempting to predict each selection and then making that selection available for use in future predictions. For large data sets, a simpler approach is to randomly sample test users, randomly sample the time just prior to a user action, hide all selections (of all users) after that instant, and then attempt to recommend items to that user. This protocol requires changing the set of given information prior to each recommendation, which can still be computationally quite expensive. An even cheaper alternative is to sample a set of test users, then sample a single test time and hide all items after the sampled test time for each test user. This simulates a situation where the recommender system is built as of the test time, and then makes recommendations without taking into account any new data arriving after 45
the test time. Another alternative is to sample a test time for each test user, and hide the test user‟s items after that time, without maintaining time consistency across users. This effectively assumes that what is important is the sequence in which items are selected not the absolute times when the selections are made. A final alternative is to ignore time. We would sample a set of test users, then sample the number na of items to hide for each user a, then sample na items to hide. This assumes that the temporal aspects of user selections are unimportant. All three of the latter alternatives partition the data into a single training set and a single test set. It is important to select an alternative that is most appropriate for the domain and task of interest, given the constraints, rather than the most convenient one. A common protocol used in many research papers is to use a fixed number of known items or a fixed number of hidden items per test user (so called “given n” or “all but n” protocols). This protocol is useful for diagnosing algorithms and identifying in which cases they work best. However, when we wish to make decisions on the algorithm that we will use in our application, we must ask ourselves whether we are truly interested in presenting recommendations only for users who have rated exactly n items, or are expected to rate exactly n items more. If that is not the case, then results computed using these protocols have biases that make them unreliable in predicting the performance of the algorithms online. Prediction accuracy is one of the most exploited properties of the recommendation systems in the process of their evaluation. In almost all of the recommender systems their accuracy is based on the quality of their prediction engine. This engine may predict user opinions over items (e.g. ratings of tourist object) or the probability of usage (e.g. visits). One basic assumption in a recommender system is that a system that provides more accurate predictions will be preferred by the user. Thus, many researchers set out to find algorithms that provide better predictions. Prediction accuracy is typically independent from the user interface, and thus can be measured in an offline experiment. Measuring prediction accuracy in a user study measures the accuracy given a recommendation. This is a different concept 46
from the prediction of user behavior without recommendations, and it is closer to the true accuracy in the real system. In the following we will discuss two broad classes of prediction accuracy measures: measuring the accuracy of ratings predictions and measuring the accuracy of usage predictions. Root Mean Squared Error (RMSE) is perhaps the most popular metric used in evaluating accuracy of predicted ratings. The system generates predicted ratings
r ui
for a test set T of user-item pairs (u,i) for which the true ratings rui are known. Typically, rui are known because they are hidden in an offline experiment, or because they were obtained through a user study or online experiment. The RMSE between the predicted and actual ratings is given by (10):
RMSE
1 T
(r
ui rui )
2
(10)
( u ,i )T
Mean Absolute Error (MAE) is a popular alternative, given by (11):
MAE
1 T
( u ,i )T
r ui rui
(11)
Compared to MAE, RMSE disproportionably penalizes large errors, so that, given a test set with four hidden items RMSE would prefer a system that makes an error of 2 on three ratings and 0 on the fourth to one that makes an error of 3 on one rating and 0 on all three others, while MAE would prefer the second system. Normalized RMSE (NMRSE) and Normalized MAE (NMAE) are versions of RMSE and MAE that have been normalized by the range of the ratings (i.e. rmax −rmin). Since they are simply scaled versions of RMSE and MAE, the resulting
47
ranking of algorithms is the same as the ranking given by the un-normalized measures. Average RMSE and Average MAE adjust for unbalanced test sets. For example, if the test set has an unbalanced distribution of items, the RMSE or MAE obtained from it might be heavily influenced by the error on a few very frequent items. If we need a measure that is representative of the prediction error on any item, it is preferable to compute MAE or RMSE separately for each item and then take the average over all items. Similarly, one can compute a per-user average RMSE or MAE if the test set has an unbalanced user distribution and we wish to understand the prediction error a randomly drawn user might face. As the prediction accuracy of a recommendation system, especially in collaborative filtering systems, in many cases grows with the amount of data, some algorithms may provide recommendations with high quality, but only for a small portion of the items where they have huge amounts of data. Most commonly, the term coverage refers to the proportion of items that the recommendation system can recommend. This is often referred to as catalog coverage. The simplest measure of catalog coverage is the percentage of all items that can ever be recommended. This measure can be computed in many cases directly given the algorithm and the input dataset. A more useful measure is the percentage of all items that are recommended to users during an experiment, either offline, online, or a user study. In some cases it may be desirable to weight the items, for example, by their popularity or utility. Then, we may agree not to be able to recommend some items which are very rarely used anyhow, but ignoring high profile items may be less tolerable. Coverage can also be the proportion of users or user interactions for which the system can recommend items. In many applications the recommender may not provide recommendations for some users due to, e.g. low confidence in the accuracy of predictions for that user. In such cases we may prefer recommenders that can provide recommendations for a wider range of users. Clearly, such recommenders should be evaluated on the tradeoff between coverage and accuracy. 48
Coverage here can be measured by the richness of the user profile required to make a recommendation. For example, in the collaborative filtering case this could be measured as the number of items that a user must rate before receiving recommendations. This measurement can be typically evaluated in offline experiments. Another, well known problem related to the recommender systems, is the so called cold start problem - the performance of the system on new items and on new users. Cold start can be considered as a sub problem of coverage, because it measures the system coverage over a specific set of items and users. In addition to measuring how large the pool of cold start items or users is, it may also be important to measure system accuracy for these users and items. Focusing on cold start items, we can use a threshold to decide on the set of cold items. For example, we can decide that cold items are only items with no ratings or usage evidence (Schein et al., 2002), or items that exist in the system for less than a certain amount of time (e.g., a day), or items that have less than a predefined evidence amount (e.g., less than 10 ratings). Perhaps a more generic way is to consider the “coldness” of an item using either the amount of time it exists in the system or the amount of data gathered for it. Then, we can credit the system more for properly predicting colder items, and less for the hot items that are predicted. We use dataset from proprietary database collected by the mixed research group composed of researchers from Faculties of Computer Science and Tourism and Business logistics at the “Goce Delcev” University. It contains 56320 ratings from 483 users for 818 tourist objects. Each user has rated at least 30 objects, and each object has been rated at least once. In order to measure recommendation accuracy more precisely, informationretrieval classification metrics was used, which evaluates the capacity of the recommender system in suggesting a list of appropriate objects to the user. With such metrics it is possible to measure the probability that the recommender system takes a correct or incorrect decision about the user interest for an item. 49
When
using
classification
metrics,
among
four
different
kinds
of
recommendations can be distinguished (Table 3).
Table 3. Classification of possible result - recommendation of object to user Recommended
Non recommended
Interesting
True-positive (TP)
False-negative (FN)
Uninteresting
False-positive (FP)
True-negative (TN)
If the system suggests an interesting tourist object to the user we have a true positive (TP), otherwise the object is uninteresting and we have a false positive (FP). If the system does not suggest an interesting tourist object we have a false negative (FN). If the system does not suggest an object uninteresting for the user, we have a true negative (TN). The most popular classification accuracy metrics are the recall and the precision. These metrics can be calculated by counting the number of test object that fall into each cell in the table and according to formulas (12) and (13).
Pr ecision
TP TP FP
Re call (True Positive Rate )
(12)
TP TP FN
(13)
Recall measures the percentage of interesting objects suggested to the users, with respect to the total number of interesting objects, while precision measures the percentage of interesting objects suggested to the users, with respect to the total number of suggested objects. In order to understand the global quality of a recommender system, we may combine recall and precision by means of the Fmeasure (14).
50
F measure
2 recall precision recall precision
(14)
These metrics are used in evaluating the quality of the recommendation. To evaluate the system a methodology which uses the k-fold and the leave-one-out together with classification metrics recall and precision were used. According to the k-fold, users in the dataset are partitioned into k parts: k - 1 parts are used to construct the model, the remaining part represents the testing set. The model created with the k - 1 partitions is tested on the remaining partition by means of the following algorithm: Step 1: One user in the testing set is selected (the active user). Step 2: One rated tourist object (the test object) is removed from the profile of the active user. Step 3: An order list of recommended tourist objects is generated. Step 4: If the test item is in the top-3 categories of recommended objects, either the true positive or false positive counter is incremented, depending whether the user liked or disliked the test item. Two distinct user groups were considered. Group A contained all users who have rated 30-60 objects (the few raters user group), while Group B contained all users who have rated 61-100 objects (the moderate raters user group). Step 1 of the proposed algorithm was repeated for all users in both groups. Steps 2-4 were repeated for all objects rated by active user. In order to understand if a user likes or dislikes a rated tourist object, a supposition is made that an object is interesting for the user if it satisfies the two following conditions (15):
Ratei , j 3 Ratei , j Ratei
(15)
51
where Ratei,j is the rate given by the user i for the tourist object j and
Ratei
is the mean
of ratings for user i. The first constraint reflects the absolute meaning of the rating scale, while the second the user bias. If a rating does not satisfy condition defined in (15), we assume the item is not interesting for the user. Once computed recall and precision, we synthesize them with the f-measure, as defined in (14). Upon conducted evaluation, the results for system precision, recall and f-measure were averaged for each group (Table 4). According to the obtained results, the developed national tourism web portal with its collaborative recommender system seems to be robust as it achieves good results in both scenarios (users with few and moderate ratings). It also accomplishes a good trade-off between precision and recall, a basic requirement for all recommender systems. Experimental results show that the proposed approach can provide satisfactory performance even in a sparse dataset.
Table 4. Average values for recommendation system precision, recall and f-measure Group
Precision
Recall (%)
F-measure (%)
Group A 75.14
79.18
77,11
Group B 81.74
85.32
83.49
(%)
Moreover, to evaluate the effectiveness of the proposed methodology in alleviating the data sparsity problem, we have checked its performance in terms of coverage with the increase in sparsity level. In this respect, Figure 9 shows the predictive accuracy of the algorithm i.e. the MAE expressed with respect to the neighborhood size. As it may be observed from the previous figure, the MAE improves, as the neighborhood size increases but, it reaches a stable performance around 90 neighbors and any further increment makes no better or even worse results.
52
Figure 9. MAE expressed with respect to the neighborhood size
Therefore, to evaluate the performance of the algorithm in the following experiments we fixed the neighborhood size to 90 and performed the experiments with different sparsity levels. Figure 10 shows how the MAE changes with respect to varying sparsity levels.
Figure 10. MAE in function of varying sparsity levels
What can be seen from Figure 10 is the impact of sparse datasets on the predictive accuracy. Furthermore, one may conclude that it performs as expected i.e. the predictive accuracy decreases as the sparsity level increases.
53
7. NECESSITY OF DEVELOPING TOURISM RECOMMENDER IN MACEDONIA Past, current and future economic indicators related to tourism development in Macedonia are already discussed in full details in Section 2. So, one may confirm the thesis that Macedonia identified tourism as a strategic priority for national economic development (Government of Macedonia, 2009). Since the inevitable relationship between tourism and ICT, one may note positive trends in usage of ICT in house and individuals. Namely, the official data point to the fact that in 2012 58.3% of the households had access to the Internet at home, which is 3.3 percentage points more compared to 2011. Additionally, one may note that 57.5% of total population in Macedonia used the Internet, out of which 62.9% have Internet access at home and live in urban settlements. Computer users among total population (being aged 15-74) have ever used a computer in 64.0% versus to similar fact of 61.5% being registered as Internet users that have ever used the Internet. It is interesting to note that 76.5 % of the registered Internet users in 2012 have a frequency of every day Internet usage. In this respect, regarding the activities on the Internet, one may note that 72.8% use on-line social networks (State Statistical Office, 2012: 2-6). So, this interesting fact represents solid starting point to urge the necessity of creating recommending system in tourism developed on the bases of an on-line social network, like: Facebook, MySpace, Twitter, Friendster, Bebo etc. In order to contribute to initializing positive trends in tourism domain, in 2008 Macedonian government established the Agency for Tourism Support and Promotion with a main mission to promote tourism resources and capacities particularly at international level. The same year, the government launched the first multiple stage promotion campaign entitled “Macedonia - Timeless” with videos broadcasted on the world renowned media CNN. Due to fact that the structure of international tourism demand is generally comprised of neighboring countries, the main accent is put on promoting the country on regional international tourism fairs, like: Vakantiebeurs in Utrecht, Holland; FERRIEN Messe in Wien, Austria; MATKA in Helsinki, Finland; 54
ITM in Warsaw, Poland; EMITT in Istanbul, Turkey; Holiday Fair in Brussels, Belgium; Vit in Milano, Italy; Holiday & Spa Expo in Sofia, Bulgaria; IFT in Belgrade, Serbia; ITB in Berlin, Germany; Mitt in Moscow, Russia; TUR in Goteborg, Sweden etc. However, the budget expenditures allocated for the implementation of the Program for tourism promotion are very modest, despite their constant increase every year. For instance, approximately €100000 were scheduled for tourism promotion in 2005 (Government of Republic of Macedonia, 2009), to €130000 in 2012 (Government of Macedonia, 2012). The need for more efforts in the field of tourism promotion in Macedonia is illustrated by the fact that Macedonia has been ranked low on the list of the most attractive destinations for travel and tourism, issued by the World Economic Forum. For example, in 2007 Macedonia was ranked as 83 rd out of 124 countries. In 2008, it was placed at the same position, but this time out of 130 countries. In 2009, a small progress was made, i.e. Macedonia was ranked 80 th out of 133 countries (Blanke and Chiesa, 2009: 31). Finally, a small progress was made in 2011, when Macedonia was ranked at the 76th place out of 139 countries. However, it should be mentioned that the majority of the countries in the region are significantly better positioned than Macedonia: Slovenia - 33rd place, Croatia - 34th place, Montenegro - 36th place, Bulgaria - 48th place and Albania - 71st place (Blanke and Chiesa, 2011: xv). Concerning the neighboring countries, only Serbia, and Bosnia and Herzegovina are ranked lower than Macedonia. In order to strengthen tourism competitiveness of Macedonia, the first national web tourism portal (www.exploringmacedonia.com) was created in 2005 as a publicprivate partnership between an international donor and the Ministry of economy. In this respect, several other private initiatives act as additional tourism portals, thus supporting
country‟s
tourism
profile,
like:
www.travel2macedonia.com,
www.macedoniatravel.com, www.go2mace donia.com, www.simplymacedonia.com, www.macedonialovesyou.com.mk,
www.macedonialovesyou.mk,
www.mystical
macedonia.com, www.macedonia-timeless.com, www.macedoniabookinghotels.mk, www.macedoniaguide.mk, etc. As of the end of 2012, the government initiated 55
creation of first electronic on-line shops (www.macedonian-handicrafts.mk) which, besides strengthening the competitiveness of small and medium sized enterprises, will contribute to enhance poorly-developed tourism supply. Despite the existence of variety and most probably, sufficient number of webportals that promote Macedonia as attractive tourist destination, so far none of them has acted as a tourism recommender. The forth mentioned advantages produced by recommenders fully justify the urgency and necessity of their design in Macedonia. What specifically leads to that fact that they assist tourists and visitors in planning and creating their trip and holiday in more sophisticated way.
8. CONCLUSION AND FUTURE WORK The objective of this study was to present and elaborate necessity of introducing recommenders in tourism, by emphasizing the case of Macedonia. In the first line, the study gave a brief presentation on the past, current and future challenges towards tourism development on world, regional and national level. It pointed out the fact that tourism was faced with many different events, some being driven by emerging markets and rapid advances in technology, particularly in digital and social media, but also by the economic environment. The research underlines the conclusion that from variety of negative challenges, the global financial crisis had the most destructive impacts thus infecting all travel and tourism-related areas. The presence of negative effects and shocks on tourism flows in Macedonia was concluded by the in-depth analysis of tourist arrivals, GDP, tourism employees and balance of payments. However, despite the shocks and misbalances, tourism has experienced continuing expansion and diversification and has become one of the largest and fastest growing industries in the world. Future challenges note positive trends on world and regional level. On local level, when referring Macedonia, one may conclude that the forecasts are much more optimistic than the actual outcomes. Furthermore, the research makes an attempt to justify the importance of introducing recommendation systems in tourism and for travel purposes. So, this 56
exploratory study results into proposing and developing a web-based platform for enhancing tourism development of Macedonia. Namely, this investigation underlines several times that tourism is defined as one of the most economically-oriented industries in the world due to fact that enhances and strengthens national economies. So, this exploration strongly argues the necessity of creating such a software module that may contribute to tourism promotion of particular destination. Moreover, the research outcome emphasizes the contribution of such platform in increasing the awareness of tourist destination that is capable of fulfilling tourists‟ and travelers‟ preferences, and respectfully in raising net tourism income. Consequently, the final tailor-made product is the web-site named “MyTravelPal”. It is highly endorsed to apply this module which is based on the method of collaborative filtering. In case of acting as a national tourism web portal it may be introduced as a tool for assisting tourists in identification of ideal and tailormade holiday place. Although it is in its initial phase, it notes accurate recommendations and guidelines and is in the line of supporting the national economy through improvement of tourism supply in more qualitative manner. Due to fact that this portal indicates the motives, preferences and reasons for traveling to Macedonia, it may be of high importance to all key-tourism actors in the process of identifying
measures
and
implementing
activities
necessary
for
creating
comprehensive tourism developmental and promotional policy. The outcomes of this study complement the optimistic forecasts for tourism demand in Macedonia being additionally supplemented with the fact of continuously increase of user ratings. Furthermore, one may underline the upward trend of web portal users which complements the positive general conclusion referring positive impulses to tourism income in Macedonia. Furthermore, the study was limited by several factors that may be addressed in some future research, like: sample size, the fragile tourism nature, limited secondary data etc. Yet, the discussed results and findings should be interpreted as selected samples to underline the usefulness of the proposed approach in contribution to tourism development and setting comprehensive tourism policy. So definitely, the 57
future work may include additional insights on improvement of presented web-based platform. Despite the forth mentioned notes, the study is rich on useful findings and poses some valuable directions for further research.
9. REFERENCES Blanke, J., & Chiesa, T. (2009). The Travel & Tourism Competitiveness Report 2009: Managing in a Time of Turbulence. Geneva: World Economic Forum. Blanke, J., & Chiesa, T. (2011). The Travel & Tourism Competitiveness Report 2011: Beyond the Downturn. Geneva: World Economic Forum. Buhalis, D. & O‟Connor, P. (2005). Information Communication Technology Revolutionizing Tourism. Tourism recreation research, 30(3), 7-16. Buhalis, D. & Law, R. (2008). Progress in information technology and tourism management: 20 years on and 10 years after the Internet - the state of eTourism research. Tourism Management, 29, 609-623. Cohen, E. (1972). Towards a Sociology of International Tourism. Social Research, 39, 164-182. Connolly, D. J., & Lee, S. (2006). Developing information technology proficiencies and fluency in hospitality students. Journal of Hospitality & Tourism Education, 18(3), 15-29. Charou, E., Kabassi, K., Martinis, A. & Stefouli, M. (2010). Integrating multimedia GIS technologies in a recommendation system for geo-tourism. In: Tsihrintzis, G. A. and Jain, L. C. (Eds.). Multimedia Services in Intelligent Environment, Springen-Verlag Berlin, 63-74. Daramola, O. J., Adigun, M. O., Ayo, C. K. & Olugbara, O. O. (2010). Improving the dependability of destination recommendations using information on social aspects. Tourismos: an International Multidisciplinary Journal of Tourism, 5(1), 13-34. Dwyer, L., & Kim, C. (2003). Destination Competitiveness. Determinants and Indicators. Current Issues in Tourism, 6 (5), 369-414. 58
Dwyer, L., Forsyth, P., & Rao, P. (2000). The Price Competitiveness of Travel and Tourism: A Comparison of 19 Destinations. Tourism Management, 21 (1), 9-22. Enright, M. J., & Newton, J. (2004). Tourism Destination Competitiveness: A Quantitative Approach. Tourism Management, 25 (6), 777-788. European Commission – Eurostat. (2007). Regional and Urban Statistics. Frew, A. J. (2000). A critical analysis of tourism information technology research. In D. R. Fesenmaier, S. Klein, & D. Buhalis (Eds.), Information and communication technologies in tourism (pp. 39-52). New York: Springer-Verlag. Gibson, H. & Yiannakis, A. (2002). Tourist roles: Needs and the life course. Annals of Tourism Research, 29(2), 358-383. Goeldner, C. R., Ritchie, J. R. B. & McIntosh, R., W. (2000). Tourism: Principles, Practices, Philosophies, John Wiley & Sons. Goossen, M., Meeuwsen, H., Franke, J. & Kuyper, M. (2009). My Ideal Tourism Destination: Personalized Destination Recommendation System Combining Individual Preferences and GIS Data. Journal of Information Technology & Tourism, 11(1), 17-30. Gretzel, U., Hyan, K. Y. & Purifoy, M. (2007). Online Travel Review Study: Role & Impact of Online Travel Reviews. Government of Republic of Macedonia. (2009). National Strategy for Tourism Development 2009-2013, Skopje. Government of the Republic of Macedonia, Ministry of Economy. (2010). Program for 2011, Skopje. Hennessey, S. M., MacDonald, R. & MacEachern, M. (2008). A Framework for Understanding Golfing Visitors to a Destination. Journal of Sport & Tourism, 13(1), 5-35. Hinze, A., Voisard, A. & Buchanan, G. (2009). TIP: Personalizing Information Delivery in a Tourist Information System. Journal of Information Technology & Tourism, 11(3), 247-264. Horwath Consulting Zagreb. (1999). Horwath Hotel Industry Survey Croatia, Zagreb. 59
Hu, Y., & Ritchie, J. R. B. (1993). Measuring Destination Attractiveness: A Contextual Approach, Journal of Travel Research, 32(2), 25-34. Jannach, D. (2006). Finding preferred query relaxations in content-based recommenders. IEEE Intelligent Systems Conference, Westminster, UK, 355360. Kabassi, K. (2010). Personalizing recommendations for tourists. Telematics and Informatics, 27(1), 51-66. Kazienko, P. & Kolodziejski, P. (2006). Personalized Integration Recommendation Methods for E-commerce. International Journal of Computer & Applications, 3(3), 12-26. Kim, W. G., & Ham, S. (2007). The impact of information technology implementation on service quality in the hotel industry. Information Technology in Hospitality, 4(4), 143-151. Kirk, D., & Pine, R. (1998). Research in hospitality systems and technology. International Journal of Hospitality Management, 17(2), 203-217. Kluge, E. A. (1996). A literature review of information technology in the hospitality curriculum. Hospitality Research Journal, 19(4), 45-64. Leiper, N. (1979). The Framework of Tourism: Towards a Definition of Tourism, Tourist and the Tourist Industry. Annals of Tourism Research, 6(4), 390-407. Law, R., Leung, R. & Buhalis, D. (2009). Information technology application in hospitality and tourism: a review of publications from 2005 to 2007. Journal of Travel and Tourism Marketing, 26: 5-6, 599-623. Leung, R., & Law, R. (2005). An analysis of information technology publications in leading hospitality journals. FIU Hospitality Review, 23(2), 55-65. Loh, S., Lorenzi,F., Saldaña, R. & Licthnow, D. (2004). A Tourism Recommender System Based on Collaboration and Text Analysis. Information Technology & Tourism, 6, 157-165. Mazanec, J. A., Wöbel, K., & Zins, A. H. (2007). Tourism Destination Competitiveness: From Definition to Explanation? Journal of Travel Research, 46, 86-96. 60
Mills, J. & Law, R. (2004). Handbook of Consumer Behaviour, Tourism and the Internet, New York: Harworth Hospitality Press. Mirzadeh, N., Ricci, F. & Bansal, M. (2004). Supporting user query relaxation in a recommender system. 5th International Conference on E-Commerce and Web Technologies (EC-Web), Zaragoza, Spain, 31-40. Morrison, A., Su, J., O‟Leary, J. & Cai, L. (2001). Predicting usage of the Internet for travel bookings: An exploratory study. Information Technology and Tourism, 4(1), 15-30. McSherry, D. (2005). Retrieval failure and recovery in recommender systems. Artificial Intelligence Review, 24, 319-338. Niaraki, A. S. & Kim, K. (2009). Ontology based personalized route planning system using a multi criteria decision making approach. Expert Systems with Applications, 36, 2250-2259. Niininen, O., Buhalis, C., & March, R. (2007). Customer empowerment in tourism through consumer centric marketing (CCM). Qualitative Market Research, 10(3), 265-282. Orio N. (2006). Music Retrieval: A tutorial and Review. Foundations and Trends in Information Retrieval, 1(1), 1-96. O‟Connor, P., & Murphy, J. (2004). Research on information technology in the hospitality industry. International Journal of Hospitality Management, 23(5), 473-484. Pearce, P. (1982). The social psychology of tourist behavior. New York: Pergamon. Petrevska, B. (2010). Tourism in the global development strategy of Macedonia: Economic perspectives. UTMS Journal of Economics, 2(1), 101-108. Petrevska, B. (2011). Economic Planning of Tourism Demand in Macedonia. Economic Development, 3/2010, 133-145. Petrevska, B. (2012a). Forecasting International Tourism Demand: the Evidence of Macedonia, UTMS Journal of Economics, 3(1), 45-55.
61
Petrevska, B. (2012b). Tourism in Macedonia - Before and After the Crisis, Conference Proceedings, International Conference “Faces of the Crisis”, Skopje, Macedonia, 63-70. Palanivel, K. & Siavkumar, R. (2010). Fuzzy multicriteria decision-making approach for collaborative recommender systems. International Journal of Computer Theory and Engineering, 2(1), 57-63. Pan, B., Litvin, S. W. & O‟Donnell, T. E. (2007). Understanding accommodation search query formulation: The first step in putting „heads in beds‟. Journal of Vacation Marketing, 13(4), 371-381. Pan, B., Xiang, Z., Fesenmaier, D. R. & Law, R. 2011. The dynamics of search engine marketing for tourist destinations. Journal of Travel Research, 50(4), 365377. Prahalad, C. K. & Ramaswamy, V. (2000). Co-opting Customer Competence. Harvard Business Review, 78(1), 79–87. Ricci, F. & Del Missier, F. (2004). Supporting Travel Decision Making through Personalized Recommendation. In: Clare-Marie Karat, Jan Blom, and John Karat (eds.), Designing Personalized User Experiences for eCommerce, Kluwer Academic Publisher, 221-251. Ricci, F. & Werthner, H. (2002). Case base querying for travel planning recommendation. Information Technology & Tourism, 4(3/4), 215-226. Ricci, F., Arslan, B., Mirzadeh, N. & Venturini, A. (2002). ITR: A case-based travel advisory system. 6th European Conference on Advances in Case-Based Reasoning, 613-627. Ritchie, J. R. B., Crouch, G. I., & Hudson, S. (2001). Developing Operational Measures for the Components of a Destination Competitiveness and Sustainability Model: Consumer versus Managerial Perspectives. In J. A. Mazanec (Ed.), Consumer Psychology of Tourism, Hospitality and Leisure, Wallingford: CABI Publishing, 1-17.
62
Setten M. V. (2005), Supporting people in finding information. Hybrid recommender systems and goal-based structuring. Telematica Instituut Fundamental Research Series, 016, Enschede, the Netherlands: Telematica Instituut. Singh, A. J., & Kasavana, M. L. (2005). The impact of information technology on future management of lodging operations: A Delphi study to predict key technological events in 2007 and 2027. Tourism & Hospitality Research, 6(1), 24-37. Singh, A. J., Kim, H., & Huh, C. (2006). Differential impacts of information technology services in the Korean hotel industry: A study of management perceptions. FIU Hospitality Review, 24(2), 78-89. Schein, A. I., Popescul A., Ungar, L. H. & Pennock, D. M. (2002). Methods and metrics for cold-start recommendations. In: SIGIR ‟02: Proceedings of the 25th annual international ACM SIGIR conference on Research and development in information retrieval, 253-260 (doi: http://doi.acm.org/10.1145/564376.564421). Song, H., & Witt, S. F. (2000). Tourism Demand Modeling and Forecasting, Modern Economic Approaches. Amsterdam. State Statistical Office. (2008). Statistical Review: Transport, Tourism and Other Services. Skopje. State Statistical Office (2012). News Release on Information Society, Nr. 8.1.12.28, Skopje. Thompson, C., Göker, M. & Langley, P. (2004). A personalized system for conversational recommendations. Journal of Artificial Intelligence Research, 21, 393-428. UNWTO. (2012a). World Tourism Barometer, Statistical Annex, 12, Madrid. UNWTO. (2012b). Tourism Highlights, Madrid. USAID (2006). Evaluation Report for Macedonian Competitiveness Activity. Skopje. Uysal, M. (1998). The Determinants of Tourism Demand: A Theoretical Perspective, The Economic of the Tourist Industry. London: Routledge. Vansteenwegen, P. & Wouter, S. (2001). Trip Planning Functionalities: State of the Art and Future. Information Technology & Tourism, 12(4), 305-315. 63
Wallace, M., Maglogiannis, I., Karpouzis, K., Kormentzas, G. & Kollias, S. (2003). Intelligent onestop-shop travel recommendations using an adaptive neural network and clustering of history. Information Technology & Tourism, 6, 181193. Wang, J. (2008). Improving decision‐making practices through information filtering. International Journal of Information and Decision Sciences, 1(1), 1‐4. Weber, K. & Roehl, W. S. (1999). Profiling People Searching for and Purchasing Travel Products on the World Wide Web. Journal of Travel Research, 37(3), 291-298. Winata, L., & Mia, L. (2005). Information technology and the performance effect of managers‟ participation in budgeting: Evidence from the hotel industry. International Journal of Hospitality Management, 24(1), 21-39. WTTC. (2009). Travel & Tourism Economic Impact - Macedonia 2009. WTTC. (2009b). Travel & Tourism Economic Impact - Central and Eastern Europe 2009. WTTC. (2011). Travel & Tourism Economic Impact - Macedonia 2011. WTTC. (2012). Benchmarking Travel & Tourism‟s economic impact against other sectors - Executive Summary, April 2012. Xiang, Z. & Pan, B. (2010). Travel queries on cities in the United States: Implications for search engine marketing for tourist destinations. Tourism Management, 32(1), 88-97. Yiannakis, A. & Gibson, H. (1992). Roles Tourist Play. Annals of Tourism Research, 19, 287-303. Yfantidou, G., Costa, G. & Michalopoulos, M. (2008). Tourist roles, gender and age in Greece: a study of tourists in Greece. International Journal of Sport Management Recreation & Tourism, 1, 14-30. Zanker, M., Fuchs, M., Höpken, W., Tuta, M. & Müller, N. (2008). Evaluating Recommender Systems in Tourism - A Case Study from Austria. In: P. O'Connor et al. (Hrsg.): Information and Communication Technologies in Tourism, Proceedings ENTER 2008, Springer, 24-34. 64
Zhang, L., Pan, B., Smith, W. & Li, X. (2009). Travelers‟ Use of Online Reviews and Recommendations: A Qualitative Study. Journal of Information Technology & Tourism, 11(2), 157-167.
65
