JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.1 (1-18)
Science of Computer Programming ••• (••••) •••–•••
Contents lists available at ScienceDirect
Science of Computer Programming www.elsevier.com/locate/scico
SMORE: Towards a semantic modeling for knowledge representation on social media Daniel Villanueva ∗ , Israel González-Carrasco, J.L. López-Cuadrado, Nora Lado Computer Science Department, Universidad Carlos III de Madrid, Spain
a r t i c l e
i n f o
Article history: Received 30 April 2015 Received in revised form 30 May 2015 Accepted 18 June 2015 Available online xxxx Keywords: Semantic data models Social media Semantic representation
a b s t r a c t This research presents SMORE, a semantic model for knowledge representation on social media. In order to provide recommendations, the model provides the elements for representing the content through the use of an ontological model and semantic techniques for the characterization and relationships between user profiles, products and social networks. In fact, with this model could be the basis of recommendation system based on social media data, and it could be exploited use by the recommendations on different products, which are stored in the Web and with similar characteristics between them. Moreover, SMORE represent the information related to user of the social networks in order to have a user characterization to be used for future recommendations in several domains. The semantic model has been evaluated with semantic data extracted from a trusted social network as, Twitter, obtaining the information specified by an expert in the field of marketing for recommendations in the automotive industry. © 2015 Elsevier B.V. All rights reserved.
1. Introduction The success of the Semantic Web, as the next generation of Web technology, can have a deep impact on the semantic modeling of knowledge, the reuse of software and thus to improve software engineering [1]. Therefore, the semantic models allow both the software engineers and machines to understand the content over of formal models and support more effective software design in terms of understanding. To achieve the full potential of the Semantic Web in formal software development, it is crucial the effective definition of proper semantic metadata for formal software models and their related software artefacts. Moreover, semantic models provide a consistent and reliable basis that can be used to confront the challenges related to the organization, manipulation and visualization of data and knowledge, in addition to play a crucial role as the technological basis in the development of a large number of computational intelligence systems. Furthermore, these models use different technologies based on new techniques from various disciplines within Computer Science domain, such as Knowledge Engineering, Natural Language Processing, Artificial Intelligence, Databases, Software Agents, etc. Besides, the methods and tools developed and integrated for this purpose have a very large application potential in many fields such as Information Retrieval, Semantic Searches, Information Integration, Information Interoperability, Bioinformatics, eHealth, eLearning, Software Engineering, e-commerce, e-government, Social Networks, etc. [2]. The social networking phenomenon has revolutionized our concept of classical social relationships and our investment in spare time [3]. Besides, social media are of great importance in influencing all types of actors and environments with
*
Corresponding author. E-mail address:
[email protected] (D. Villanueva).
http://dx.doi.org/10.1016/j.scico.2015.06.008 0167-6423/© 2015 Elsevier B.V. All rights reserved.
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.2 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
2
a public that is increasingly participatory and which demands, ever more intensely, the finding of various contents from different paradigms [4]. At the same time, social media provide great enrichment, as they offer the tools for a new relationship subject to a largely globalized and demanding society. Thus, the information can be used for recommendations. Social media and Recommender Systems (RS) have emerged as one of the interesting research topics in recent times [5] and [6]. Over the years the definition of RS has evolved the advances of new technologies which are continually emerging [6]. RS have become extremely common in recent years, and are applied in a variety of applications. RS are software tools and techniques which providing suggestions over items to be of use to a user [7]. These systems use a specific type of information-filtering techniques that attempt to suggest various contents [8]. Prominent examples include sites, such as Flickr1 and YouTube,2 for sharing pictures and videos, blog and wiki systems such as Blogger, social tagging sites such as Delicious,3 social network sites such as MySpace4 and Facebook,5 and micro-blogging sites such as Twitter.6 The latter is a service that allows users to send plain text messages, called tweets, of short length, with a maximum of 140 characters which show the user information in a home page that is likely to interest the several users on the Web. Moreover, RS use tools and techniques that provide suggestions for various products to be of use to the users [9]. Additionally, they combine ideas from user profiling, information filtering and machine learning to deliver users a more intelligent and proactive information service through concrete product or service recommendations that match their preferences and needs [10]. RS play an important role in such several and highly-rated Internet domains as Amazon,7 Netflix,8 YouTube, iTunes,9 Tripadvisor,10 Lasf.fm,11 and IMDb.12 Thus, many media companies are now developing and deploying Really Simple Syndication (RSS), as part of the service they provide to their subscribers. Besides, RSS is a technology for syndicating information such as the content of websites appropriate for managing several domains in the big field of Artificial Intelligence [11]. Furthermore, these systems are used for predictions on models in different products which are found on the Web [9]. The models can be used for filtering recommendations for individual-level or group-level prediction [12]. These techniques are a current trend in the development of RS and smart environments since they combine characteristics of content-based recommendations and filtering information recommendations [13]. However, these techniques, when used for content classification, generate specific models for a single domain, thus delimiting the model for their function. For example, if a user would like to decide between new releases of music, plays, movies or TV shows, with these delimited models this user would get limited options for deciding on the great diversity of contents and products offered on the Web. This paper presents a Semantic MOdeling REpresentation for extracted from information of social networks (SMORE). The model is based on semantic technologies that enable a characterized content to be represented. The model could be used by an RS that allows the generation of multi-domain recommendations on content stored on the social media platforms on the Web. In this research, the authors use the term multi-domain to refer to different or various lines of application upon the products offered on the Web. Thus, the model is not restricted to working in a single domain. 2. Related work This section presents relevant works related to the technologies of the RS. First, the importance of the application of recommendations to multi-domains. Second, the main characteristics of semantic models applied to various domains. Finally, the relevant literature on the models applied upon the social media recommendations. 2.1. Recommender Systems RS are used to recommend potentially interesting items to users in different domains; they exist to satisfy the need to provide potentially interesting assistance for users on multi-domains [14] and [6], hence the information allows items that match a user’s preference, tastes, and needs to be identified. As well as this, it has become an important research area, since the appearance of the first papers on collaborative filtering in the mid-1990s until today [15]. The RS allow for the attainment of different scenarios and environments to achieve access to information and are tools for interacting with large and complex information spaces [16]. Moreover, these systems provide a customized view of these areas, giving priority to
1 2 3 4 5 6 7 8 9 10 11 12
https://www.flickr.com/. https://www.youtube.com/. https://delicious.com/. https://myspace.com/. https://www.facebook.com/. https://twitter.com/. http://www.amazon.es/ref=gno_logo. https://signup.netflix.com/global. http://www.apple.com/es/itunes/. http://www.tripadvisor.es/. http://www.lastfm.es/. http://www.imdb.com/.
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.3 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
3
issues that may be of interest to a user. Over the years the definition of Recommendation Systems has evolved with the advancement of new technologies which are continually emerging [17]. RS are a kind of automated and sophisticated Decision Support System (DSS). They are necessary to provide a personalized solution in a brief form without going through a complicated search process [18]. As well, RS are systems devised for enabling interaction, negotiation, collaboration and also to coordinate activities in order to establish agreements and manage their efficient execution [19]. Furthermore, RS are used to give users suggestions about interesting items in different domains, such as e-commerce [20], tourism [21], movies [22], music [23], restaurants [24], TV programs [25], social games [26] or virtual worlds [27]. It is noteworthy that in the literature there are proposed solutions to classical problems of models for recommendation systems. The literature basically classifies these models into the following types [6]:
• • • • • •
Collaborative Recommender System, [28]. Demography Recommender System, [29]. Content-based Recommender System, [10]. Knowledge-based Recommender System, [30]. Hybrid Recommender System, [31]. Economic-factor-based Recommender System Economic Factor Based, [32].
The importance of these systems is that they have the ability to explain their reasoning because of their ability to provide personalized, meaningful information recommendations for different domains. In these systems it is argued that the extraction of more meaningful data are the models in that the recommendations are based, according to [33] and [6]. It is thus a challenge that must be addressed to enable the development of RS in a way that makes them more understandable, effective, and acceptable [34]. RS models are able to generate results for each user, whereby the information is personalized and more relevant because they take into account each user’s personal interests. 2.2. Semantic models for recommendation This section presents semantic models used in semantic RS in several domains. Semantic models are conceptual data models in which semantic information is included. This means that the model describes the meaning of its instances. Moreover, semantic data model is an abstraction that defines how the stored symbols (the instance data) relate to the real world. The formal semantic models have one characteristic in common, the use of profiles to represent information about the needs and long-term interests of users [35]. Accordingly, user profiles become a key filter for efficient semantic information, since an inadequate modeling profile can lead to low-quality recommendations which are irrelevant to the user [36]. Moreover, the representation of semantic knowledge in the various areas, generates, distributes and enriches the customized content: content that is constantly demanded by different kinds of information and different types of users [37] and [38]. Therefore, the conceptual models for the Knowledge-Based Systems (KBS) are a large field of research, since they allow knowledge to be modeled, managed and processed through various information filtering techniques in the field of the Semantic Web [39]. In some cases, the domain in which recommendations originate is important since it suggests the need for personalization techniques to be used, according to [40]. The semantic models for RS are fast moving in a multidisciplinary way. For example, customer relationship management [41], tourism [42], multimedia [43], travel [37], health domain [44] or [45], games [26], financial [46], social networks [47], e-commerce [48] or media [49] to cite just some of the most relevant cases. Nowadays there are models which use a set of rules, machine learning, decision trees, neural networks or a set of numerical numbers representing the corresponding weights for some specific characteristics upon the various products or user profiles and social networks that are applied in various domains on the Web proposed by [50]. For example, [51] proposed a system that calculates the confidence levels between clicked products using association rule mining. On the other hand, [52] present a methodology which is based on a variety of data mining techniques, decision tree induction and association rule mining. Hsieh, presents an expert system with based techniques of mental accounting and artificial neural networks on recommendations [53]. López-Cuadrado presents a framework based on ontologies that allows experts to represent and share their knowledge of business processes with other experts by means of shared and controlled vocabularies. The framework also allows the execution of business processes represented by experts [54]. Modern digital media technology, and the data contents abundant on the Web, make that sharing, collaborating and connecting with each both faster and easier. Besides, the information is demanding that many people at the present time either accessible. Guy et al., present the study of a model for personalized recommendations on social media elements based on people and tags, with Content Filtering (CF)–Content Based (CB) hybrid techniques in the social media domain [5]. Towle and Quinn present a Knowledge-Based model (KB) inferred from the ratings given by the users of certain products in the online shopping domain, they comment that the additional information given by the user and product models can give the system leverage in difficult recommendation tasks, and also alleviate both the “early rater” problem and the “sparse ratings” problem experienced by current RS [55]. Tiroshi et al., comment that where there is a variety of social web services, each user is unique for modeling specific characteristics, but this could aid RS in different ways [56]. Despite the fact that the
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.4 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
4
Table 1 Semantic models for recommendations. Author
Semantic
Multi-domain models
Generic
CB
KB
CF
H. Agius and M. Angelides, Ref. [57] Glass, Marx Schmidt and Sivrikaya, Ref. [58] Di Noia et al., Ref. [59] Raimond et al., Ref. [60] Fernández-Tobías et al., Ref. [61] Ruiz and Aldana, Ref. [62] Blanco-Fernández et al., Ref. [63] Tsatsou, Mezaris and Kompatsiaris, Ref. [64] Ghani and Fano, Ref. [15] Belk, Germanakos and Tsianos, Ref. [65] Hofmann, Ref. [66] SMORE
X X X X X X X X X X
X
X X X X X X X X X X X
X X
X X X X
X
X X
X
cited research can generate recommendations, all models presented in them are limited because they are only applied in one domain. Moreover, these systems offer only one recommendation of products or items when there is a large amount or variety of products in the social networks available to offer to the final users. Thus, this research is also supported in the survey given by J. Lu et al. [6]. Firstly, the authors of [6] do not provided information about the existing semantic models in the literature. Secondly, none of the models presented in said survey are generic models. Therefore, in this research the semantic models are considered as an important area of exploitation within RS. In addition, SMORE proposes a generic model for the representation of knowledge adaptable to any domain. Thus, it is selected some semantic models that used the techniques mentioned but are not generic models. Table 1 shows a comparison between semantic models applied to different domains with CF, KB and CB techniques. The analysis of semantic models presented in Table 1 shows an interesting point; all the authors describe semantic models with different techniques [59,60,62–64,66]. Moreover, the following authors [57,58,61,65] remark that the semantic models presented in their research are multi-domain because their models are applied to two domains. However, these models include a feature in common: they are not generic models. Therefore, none of the models presented in Table 1 are generic models that can be moved to another domain without changing their main concepts. But rather, are models that are applied to a single domain. In addition to this, the different modeling products can only be applied to a single specified domain, and it should be noted that the models generate good results when applied to domain recommendation. In the models there is a limited similarity, making it difficult to apply such a model to a new domain or switch to a new product without destroying their knowledge bases, limiting the vast information among domains available on social media for users. The SMORE model that is presented here can be adapted to any domain of structured and semantic content; in fact it is a generic model which enables multiple products from social media to be offered. 2.3. Social media recommendations Nowadays it is well recognized that user-generated content (e.g., product reviews, tags, forum discussions, entertainment and blogs) contains valuable user opinions that can be exploited for knowledge-based systems according to [67] and [6]. Fig. 1, shows a social media landscape and new forms of communication between companies and their customers, the new way to create a hosted social network platform, thus enabling content creation or the requirement of direct communication with consumers. The new forms of communication make their followers delve into the world of social media. Social media RS have emerged as one of the interesting research topics in recent times. These systems form a specific type of information-filtering technique that attempts to suggest information (blogs, news, music, travel plans, web pages, images, tags, content communities, collaborative projects, social networking sites, virtual game worlds, virtual social worlds) that is likely to interest the users [68]. Social media and knowledge-based systems can mutually benefit one another: on the one hand, social media introduce new types of public data and metadata [69], such as tags, ratings, comments, and explicit relationships of people, which can be utilized to enhance recommendations [2,70] and [71]; on the other hand, knowledge-based technologies can play a key role in the success of social media applications and the social web as a whole, ensuring that each user is presented with the most attractive and relevant content, on a personal level [5]. For example, Chen, Zeng, and Yuan proposed a unified framework for generating three types of recommendation [72]. The number of Internet users is increasing rapidly, and a huge number of people now interact with each other on online social networks [73]. In this way, the Web community has become similar to real-world society. [74], generated an algorithm to evaluate the responses of the general public through surveys of representative users in social networks. Increasingly there is a larger number of brands in social media. Facebook alone, an emblematic social media, has over 955 million active users who log on at least once every 30 days. Half of these active users are actually logged on every day. Therefore, the brands and different products that are offered to the users on the Web, are increasingly used by RS for different domains. For example, Laroche, Habibi, and Richard present a model to show how on social media, brand-based
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.5 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
5
Fig. 1. Social media landscape.13
communities could cement relationships between customers, marketers, product, brand, and other customers, and how these relationships could enhance brand trust and loyalty [75]. Christensen and Schiaffino present a RS for suggest entertainment to groups of users which utilize different methods to the generation of recommendations [29]. There are huge collections of heterogeneous media data on social websites, digital libraries, and mobile Internet apps. Many social behaviors can be reflected by integrating these heterogeneous media data, which also can be used to provide cross-media recommendations. Internet users, therefore, need to receive a variety of information from products of different web domains, which allow the generating systems to make recommendations on the social media. To process the information, there are models, such as ontological models or semantic models, which are based on the knowledge that applies to different domains; these models can generate more accurate knowledge about the actors and at the same time, they can cover the semantics of various contents towards a specific content, making it possible in this way to obtain a greater amount of information about the different domains and products on the Web. Social media has provided a market for RS. Thus, opportunities arise to take advantage of interactions between buyers and sellers through the web. The SMORE model can offer multiple products from social media starting from an RS that help users navigate through large information spaces on different product descriptions. 3. Knowledge representation model The semantic model proposed is based on the representation of content which is acquired from social media. For the representation of information the authors have defined concepts that restrict and determine the model. Furthermore, to publish and share data, the model chosen is the Web Ontology Language (OWL), a family of knowledge representation languages that allows users to write explicitly; a formal conceptualization of domain models [76]. In the context of knowledge management, ontology is referred to as the shared understanding of some domains, which is often conceived as a set of entities, relations, functions, axioms and instances according to [77]. The SMORE ontology allows the set of concepts that form the proposed model, its properties and its relationships to be represented. Fig. 2, shows the high-level abstract architecture for knowledge representation from social media. The abstract model represents the content management extracted from social media sources for the characterization of certain products, brands, service and profiles. A knowledge-based system, part of the generation of recommendations regarding the semantic content to provide relevant results in different domains.
13
http://www.fredcavazza.net/2008/06/09/social-media-landscape/.
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.6 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
6
Fig. 2. Abstract architecture for the knowledge representation.
Fig. 3. High-level modeling.
3.1. SMORE ontology The high-level overview of the data model is presented in Fig. 3, which consists of profiles and products characterized from user preferences and process structured content of social media. The proposed model is generic because it can be applied to different social media domains. Firstly, the concepts are described, and later the full model is presented. In the proposed ontology, a basic class hierarchy containing the elements of the system is established. The high-level overview of the data model consists of five main pillars: the modeling of a product; the preferences which are determined by a user; modeling user profiles and the social media pillar from which the information is extracted. The pillars are described below.
• • • • •
User: Refers to the actors which will be determined through an Id identification. Social Media: Social media play an important role in data extraction and creation of profiles and products characterized. Products: Products, services or brands which are stored on the Web. Profiles: Modeling different types of characterizing profiles and linked to the products from social networks. Preferences: This class determines user preferences across at rules the system established in the model.
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.7 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
7
Fig. 4. SMORE ontology.
Table 2 Description concepts of SMORE ontology. Concept
Definition
Example
User
Actor’s assets and liabilities interacting on the Web for editing, publishing and sharing information. Communication platform on the Web. Its structure allows dynamic interaction between users. A set of tangible and intangible attributes offered from a line on the market that meet a need. A set of characteristics associated with an element or category.
User@tweetero
Social Network Product Profile Social Media Classification
Types of communication platforms where content is created by using Web technologies.
Characteristics of User Characteristics of Product Characteristics of Social Network Permanent
Users can be distinguished from others depending on various characteristics.
Twitter Brands Shy, outgoing, rational, emotional Social Networking, Blogs, Microblogs, video games, photos, Wikis, Podcasts, photo and video Education, experience
Products can be distinguished from others depending on various characteristics.
Design, engine, security
Social Network can be distinguished from others depending on various characteristics.
Microblogging, text messages Introverted
Temporary Category Ratings of Product Category of Products
It is a concept which born from the user’s characteristics, thus it may or may not be permanent. It is a concept which born from the user’s characteristics and the product characteristics, thus it may or may not be temporary. Represents the content of a class associated with a profile characterized and product characterized, the content is inferred from the information stored in the social networks. Represents to content of a class associated with the preferences of a user and the characteristic product, the concept allows the ratings of a particular product across to user. It is a concept born from the associated content on the Web, there are sub-categories that can store a content associated with those categories originally described.
Geographic location, Date of Expiry Ecologist / Security Likes and preferences Consumer business
When taking a formal approach in a model context, the ontology allows management to knowledge for modeling the tastes and preferences among the profiles and products to be generated. Thus, the content is characterized by users, so that such data allow to create a new knowledge upon by the actors. Fig. 4 depicts the main concepts of the ontology designed for SMORE. In addition, with this model a characterization for semantic modeling can be performed. The relations and concepts for this operation are described in Section 3.2. The relations of SMORE ontology are detailed in Table 2.
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.8 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
8
Fig. 5. Concepts and relationships of formal model.
Fig. 6. Relationships and characteristics.
3.2. Relations in SMORE ontology Social media are a set of media for creating user content by using the technologies of the Web, so that allows to actors provide publication and the exchange of information of easy form [78]. In the SMORE model, a social network belongs to a type of classification in the social media content. In fact, a user has a concept called , this concept stores a that is a type of property that belongs to the , so that as well as serving as information stored for a profile, which is another type of property belonging to the user, the profile information is provided by an expert. In addition, at present there are vocabularies which can help to determine a user profile. An example is SentiWordNet,14 which helps to know the opinion of users through a list of positive and negative social media that can be approximated to product rating. This is based on sentiment scores across positivity, negativity and objectivity of words, according to [79] and [80]. In the formal model, the user has concepts whereby the model establishes relationships. For example the concepts called and , are both relationships that link the user with a this kind of concept is given by an expert and the product starting from the information stored on the Web. As well as this, it has a which belongs to a concept type of classification in the social media content . Therefore, those concepts described in Fig. 5, allow the characterization of information about the formal model by means of the relationships among them. Fig. 6 shows the concepts and the relationship between the characteristics of a , and . The concept called contains characteristics that may or may not be associated with a social network which in turn belongs to a user. Also, a product belongs to a , this concept allows for classifying characteristics of a given product with the interests of a user, hence the existence of the concept .
14
http://sentiwordnet.isti.cnr.it/.
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.9 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
9
Fig. 7. High-level relationships for a validation.
Fig. 7, shows the high-level relationship between the three principal concepts: , and . These concepts allow the relationship for the rating of a given product through levels of interest for the users . The validation of the product assessment is a particular characteristic in relation to the user; this concept allows a specific product to be rated by a user through the extraction of content from the social network. As mentioned, the proposed framework allows the information extracted from social networks to be modeled by classifying these concepts and relationships. Thus, a characterized content is obtained over a product and a user profile. Table 2 describes the defined concepts on SMORE ontology which through content extracted from the Web makes the characterization possible. 3.3. Framework architecture based on SMORE ontology Frameworks for the representation of semantic knowledge in multi-domain areas, generate, distribute and enrich the custom content [81]. Thus, content stored on social media is very important, and is constantly demanded by different kinds of domains and different types of user [82]. Therefore, the conceptual models for the knowledge-based systems are a large field of application and exploitation, since they allow knowledge to be modeled, managed and processed through various information-filtering techniques in the field of the Semantic Web [54]. SMORE architecture is based on a framework which allows modeling from content on social networks for knowledge representation, because as mentioned above, the first need is to model the content which will serve as a basis for the further development of the recommendations. The framework architecture based on SMORE ontology is shown in Fig. 8. The research focus is on information extraction and content modeling starting from social media. The SMORE ontology represents products and product features. These products and features are mentioned by users of social networks. The framework considers user mentions as single signal of interest by users. Thus, if a user mentioned in a tweet the word “security”, then the user would be associated with this product feature in order to obtain future related recommendations. This association is made through the ontology features and rules. It should be noted that the profile rules are determined by an expert. The product characteristics allow identified key tags to be semantically associated with the user, for which purpose nowadays there are vocabularies that help to determine these preferences, such as SentiWordNet, described in Section 3.2. The data are extracted from social media and stored in a knowledge base characterized by creating a new profile. In this process the relevant information is classified, extracted and converted to the proper format using Java technology. Simultaneously and automatically, the Jena API, a Java-based framework that enables to read, sort and modify data using OWL, is used. At this point, the population process begins including domain concepts determined by the expert in the ontology: products and their features. Then, the system continues the population process by including users of social networks that mention the key concepts to establish partnerships and user profiles by rules. In addition, the information extracted from social media is linked to that profile which is characterized across preferences. All this information is stored in the knowledge base and as a result, a product characterized across the information between a profile and a product is generated. The semantic model is populated with information extracted from social media. In addition, it is the basis for a Knowledge-based system that uses user preferences to determine a set of recommendations and suggestions about a characterized product and, in turn, the information serves to offer future recommendations about the product to a specific user through an RS. The framework uses ontological model and semantic techniques to represent the relationships between user profiles, products and social networks which, in turn, serve as a support for the generation of recommendations using an RS. In addition, SMORE allows the characterization of different products on social media to be modeled.
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.10 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
10
Fig. 8. Framework architecture based on SMORE ontology.
Hence, the model enriches the recommendations generated by the user, whereby it sets a base for a knowledge-based system from multiple domains. The content for user modeling is generated through the profiles and the information extracted from social media (see Fig. 8). The model and relationships are described in the following section. 4. Validation and results 4.1. Experiment design The semantic model proposed in this research, allows the representation of knowledge over a social network, the product and a user profile starting from the semantically information extracted from social media. In addition, the model enables a characterization through keywords upon the content from users, products and the social networks to be represented. It is noteworthy that the SMORE model will allow, from a recommendation system, the generation of recommendations on this content. Moreover, the social media information that contains the keywords defined by the expert of the domain is extracted from Twitter. In the design of experiments, all the tweets mentioned in the paper are used because to perform the filtering of information, all the tweets that are related to the keywords of the domain are extracted. To validate the proposal case study, Twittiment15 and the social network Twitter has been used. Twitter is a microblogging site that generates a constant stream of communication, some of which concerns events of general interest [83]. Hence, by posting a tweet it is possible to know the information behavior and it therefore has an affective component, in the sense of judgments or intentions according to [84]. Tweets are text messages of up to 140 characters that propose an intuitive, domain-specific, unsupervised, lexicon-based approach that estimates the level of emotional intensity contained in the text [85]. Therefore, in the SMORE model, the social network Twitter is used as a tool for analyzing tweets since it allows the users to write plain-text short-length messages, with a maximum of 140 characters, called tweets, on a variety of subjects and different domains in the Web. For this case study, a total of 1,377,713 tweets for the period from 25/06/2014 to 25/09/2014
15
http://twittiment.com/.
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.11 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
11
Fig. 9. Sample of the concepts for a case study the formal model.
were analyzed. As well as this, for this case study information concerning the characteristics associated with the automotive industry has been selected, whereby, through monitoring a set of brands, an expert determines the characteristics. On the Web, users talk about a variety of topics and products, but in this research a topic concerning the automotive industry has been selected for analysis. Thus, the users mention, through a social network, their characteristics relative to different brands of vehicles. For this case study, the following brands were selected: BMW, Audi, Fiat, Ford, Honda, Hyundai, Nissan, Peugeot and Renault. These brands were chosen to delimit content analysis as they belong to a category of products denominated vehicles, the category associated with a determined product in the automotive industry domain. Therefore, by analyzing this content it is possible to identify the characteristics of products based on the words employed by users. 4.2. Experiment results In the proposed case study, keywords and rules are provided by experts in marketing (automotive domain). Marketing experts also define the possible user profiles and the rules to incorporate a user to each profile. Moreover, the information that contains keywords defined by a domain expert has been filtered through the application of rules. Hence, in order to perform the different analysis of this research the tweets containing those keywords have been extracted from Twitter. The process is the following one: 1. The system uses words that are mentioned by users in their tweets. Thus, the system identifies the keywords associating them with the concepts defined in the ontology by means of rules. 2. The system also identifies the user who is mentioning the words. 3. The system links the user to the corresponding profile by means of the rules defined by the expert. These rules are also based on the selected words of the domain. In order to validate the SMORE ontology an experiment is carried out which enables the relationships and concepts described in Section 3.2, to be analyzed. Fig. 9 represents the formal model applied in the automotive domain. In addition, it represents a concept called Ford as a brand product type as well as the social network category type and a security characteristic. The test for knowledge representation of these concepts was carried out over the same period mentioned in Section 4.1, where the users mentioned products related to brands of vehicles and characteristics about security. Therefore, a user has a category of vehicle product, which is a Ford, associated with the automotive industry. In the same way, the user has characteristics of a social network where mentioned at security, which in this case is the social network Twitter, the definitive characteristics of the products of the domain are determined by the expert. Fig. 10(a) shows the evolution of the mentions of the various brands on Twitter given by tweets, from which are deducted relationships of the concepts described above (see Fig. 9). The concepts are associated to users, the vehicles and, in turn, with different brands. Thus through said network 1,377,713. Tweets, concerning different brands of vehicles which are mentioned by users have been analyzed. In this research, the role of the expert is to determine the characteristics of the product in the domain. Therefore, the expert could define the characteristics of the products for a particular domain in the knowledge base (ontology). Therefore, the expert could define the characteristics of the products for a particular domain in the knowledge base, but he does not analyze each tweet. In the process of extracting information, it is not performed an extraction of words, because words are defined by the expert in the domain, so that the tweet that mentions the word or key characteristic is extracted by means of rules.
JID:SCICO AID:1912 /FLA
12
[m3G; v1.159; Prn:16/07/2015; 14:05] P.12 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
Fig. 10. (a) Graphic of the products by dates, (b) Graphic of the products by brands, both obtained from Twitter. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Moreover, the expert determined the rules that relate the words and characteristics of the products defined in the ontology. Therefore, the tweets that mention the keywords according to the rules provided by the experts are extracted. Fig. 10(b) shows the cumulative number of mentions in the period studied, where it is possible to see the various brands that have been mentioned on the social network. Therefore, the relationships of the formal model support the results of the analysis, as the different vehicle brands that have been mentioned by Twitter users can be identified; in addition, the information that will serve by way of preferences for future recommendations starting from an RS is stored.
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.13 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
13
Table 3 Description of concepts the formal model. Concept
Value
User Category Of Product Characteristic Of Product Social Network Classification Of Social Network Characteristics Of User Category
user@tweetero_(fictitious_name) Vehicles Ford Twitter Microblogging Security Profile
Fig. 11. Relationships of characteristics by concepts.
Fig. 12. Graphic of the analysis that mentioned relationships of characteristics.
In this environment, Ford is the vehicles brand which the users have mentioned most, with 305,650 tweets. Table 3 describes the concepts which support the formal model of Fig. 9, which use the values obtained for this experiment. Fig. 11 shows the relationship of the concepts of the SMORE ontology for a case study, by means of which the analysis of the characteristics of a product called Audi is displayed; the product belongs to a category of vehicle type. In addition, the product has a characteristic that is called security, further wherein it is a characteristic of interest to the user. For example, users mention the word security in their tweets, in addition, the expert expresses that they have a corporate profile, because they mention an Audi brand product. Therefore, in Fig. 12 shows users with a corporate profile that mention the Audi brand and with security characteristics. In this test, 90,830 tweets are analyzed, of which 195,000 corresponding to two-percent, mention security characteristics, the model determines a trait of interest in the said product and characteristic, as such information would be used in future recommendations by means of a recommendation system. Thus, the user’s relationship with a social network allows the information for the characterization of a product to be found. The concepts establish a relationship between the user interests based on the entries made with said characteristics. These concepts have been described in Section 3.2. During this period the tweets that are related to brands and characteristics of the automotive industry have been analyzed. The period analyzed in this experiment was the same as that commented on in Section 4.1.
JID:SCICO AID:1912 /FLA
14
[m3G; v1.159; Prn:16/07/2015; 14:05] P.14 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
Fig. 13. (a) Analysis of tweets that mention Audi brand, (b) Analysis of tweets that mention security characteristic.
Fig. 14. Relationships of concepts to validation for case study.
Fig. 13(a) enables the results of the analysis of the concepts shown in Fig. 11 to be visualized. As well as this, it supports the relationship established in Fig. 11 between the concepts of characteristics, product, social network, category and users. Whereby, for this test, 248,805 tweets that mention a product called Audi were extracted, of which 734.00 mention the security characteristic. The information for this analysis is extracted from the social network Twitter. In this test the vehicle brands for the product characterization are identified. Thus, the importance of identifying the preferences of users regarding those characteristics that could be used for future recommendations is clear. Fig. 13(b) shows the extraction of information concerning the Audi brand and its security characteristics. In fact, the information allows the user’s interest in a product to be known; this is in order to analyze the content extracted from the social network Twitter that also helps us to know the preferences of users. In addition, on the basis of the semantic model, the Audi product information and its own characteristics, such as “Security”, are noted. Therefore, this characteristic allows us to identify a user’s preferences for the validation of a certain product. In high-level relationships, the social network user and product concepts are detailed. These concepts are the key that allows information obtained from Web data (see Section 3.2) to be modeled and validated. Moreover, the relations of the modeling concepts between a user who has the social network Twitter and the name of product are also represented. This concept belongs to a category of vehicle type, the product also with specific characteristics, such as price and consumption. These concepts are shown in Fig. 14. Furthermore, for this test the same period mentioned in Section 4.1, has been used. Thus, for this case study 2778 tweets have been extracted from Twitter for analysis of the relations and concepts of the model proposed. In this test, the experiment allows users’ tweets which mention the Renault Clio product, and its price and consumption characteristics, to be known. The user also has a relationship with the valuation of these characteristics. In order to meet users’ interests, and so that in the future the information will serve to make recommendations through a recommendation system, the information is stored in the SMORE ontology. Several approaches can be applied to the valuation. For example, it can be based on sentiment scores that allow users’ opinions to be known regarding their rating of products through the positivity and negativity of words by means of SentiWordNet vocabulary. Thus, this information serves to semanti-
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.15 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
15
Fig. 15. (a) Analysis of tweets that mention Renault brand, (b) hashtags users, (c) characterized product.
Fig. 16. Analysis of tweets that mention Audi, Citroen and Renault brands. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
cally characterize the brands and products, as well as it being possible to incorporate new concepts that can be used by SMORE ontology from the hashtags of the tweets. It is reflected in the relationship between the user, the rating and the Price/Consumption concepts of Fig. 14. Therefore, 104,847 tweets are analyzed in this test the users have mentioned to the Renault Clio brand and the price/consumption characteristics for the period of time set, the results of this test are shown in graph (a) of Fig. 15. In addition, graph (b) of the same figure shows the results obtained through the keywords selected and mentioned by users. Graph (c) of Fig. 15 shows the results obtained from the analysis of the tweets which are mentioned by the users. In this test, 9729 tweets regarding the Renault brand and the characteristics mentioned price and consumption are identified. In a second case study, 374,951.00 tweets were analyzed. For this test, the users mention three brands of vehicle: Audi with 248,805.00 tweets, Renault with 104,847.00 tweets and Citroen with 21,299.00 tweets. These brands belong to a product category. Fig. 16 shows a graph where users mention the brands over this established period and the percentage of the tweets in which each of the brands are mentioned is also visualized. Moreover, three types of characteristics are also mentioned. In Fig. 17(a) the users mention price and consumption in 116 tweets; design in 2465 tweets; sport in 2314 tweets and ecology in 2554 tweets. Fig. 17(b) shows another example for the analysis of technology characteristics, with 5453 mentions, design with 2465 mentions and price/consumption with 116 mentions from the users. In this section, it has been presented a case study where the SMORE model is applied for a specific domain. However, the model is based on generic concepts like “product” or “product characteristic” and they are applicable to other domains. Moreover, it can be populated with the information retrieved from different social media.
JID:SCICO AID:1912 /FLA
16
[m3G; v1.159; Prn:16/07/2015; 14:05] P.16 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
Fig. 17. Analysis of tweets that mention a different characterization of product. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
5. Conclusions and future work The semantic model presented in this work allows the representation of knowledge extracted from social media. Moreover, the concepts described in the model can generate a characterization based on content a social network, a product description and a user profile. Comparing of SMORE model with the semantic models presented in the literature (see Table 1), we found that SMORE present a difference with respect the rest of analyzed models, SMORE is a generic semantic model, because it can characterize structured content for different products and characteristics without changing the concepts of the model. In fact, SMORE can model and manage knowledge from different user profiles, products and social media characterized for different domains. Additionally, the SMORE model is valid for establishing a knowledge-base that in the future will enable the generation of recommendations for users regarding different products of several domains. To the best of our knowledge (as shown in the literature review) existing models are domain dependents. Moreover, the key differences of SMORE in comparison with the other models mentioned in the literature have been presented. The ontology-based model proposed allows the management of the extracted from social media information and, in turn, establishes the relationships between the concepts for characterization of the user profiles, products and social networks. The implications of the inclusion of the values upon the concepts of user enable the different user preferences regarding a particular product to be characterized, and in this article it has been demonstrated through the representation of the characteristics and vehicles brands on Twitter. Therefore, the model is capable of representing the content of a product extracted from the user’s preferred social network. For this case study, set of brands has been selected to determine the final characteristics starting from user preference. In addition, this model allows the semantic annotation of keywords; this information will be used as preferences for future recommendations. In the evaluation performed, authors have validated that the model supports the representation of the elements required for recommendations, connecting products and their characteristics with potential users. The information has been extracted from Twitter as a case study, and statistics presented support both concepts and relations of the proposed ontology. The conclusion is that the proposed ontology is able to represent knowledge based on products, characteristics and users extracted from social media for future recommendations. We presented a case study where is applied the SMORE model for a given domain. In this sense, in Section 5 it is described and explained with more detail why it is valid for different domains because the concepts of the ontology are general and applicable for all domains (product, product characteristics, etc.). As future work a recommendation engine will be developed on the Framework Architecture based on SMORE ontology. In this way, recommendations based on the information extracted from social networks will be generated. Accordingly, dynamic recommendations could be made based on the continuous evolution of user and product information published in social media. Moreover, in future work it would be possible to make a sentiment analysis of different products that have been mentioned by users through their tweets. The analysis would be based on the concepts of SMORE ontology and the keywords and rules that are defined by an expert in a specific domain. Additionally, the authors are working on the extension of the social media extractor in order to capture information of users and products published in other social media networks. Acknowledgements The Authors are very grateful to the General Council of Superior Technological Education of Mexico (DGEST). Additionally, this work was sponsored by the National Council of Science and Technology (CONACYT) and the Public Education Secretary
JID:SCICO AID:1912 /FLA
[m3G; v1.159; Prn:16/07/2015; 14:05] P.17 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
17
(SEP) through PROMEP and the project “FLORA: Financial Linked Open Data-based Reasoning and Management for Web Science” (TIN2011-27405). References [1] H.H. Wang, D. Damljanovic, J. Sun, An automated tool for semantic accessing to formal software models, Sci. Comput. Program. 95 (Dec. 2014) 93–111. [2] R. Valencia-García, A. Rodríguez-González, R. Colomo-Palacios, in: Special Issue on Systems Development by Means of Semantic Technologies, Sci. Comput. Program. 95 (2014) 1–2. [3] M.A. Mayer, A. Leis, A. Mayer, A. Rodriguez-Gonzalez, How medical doctors and students should use social media: a review of the main guidelines for proposing practical recommendations, Stud. Health Technol. Inform. 180 (2011) 853–857. [4] D. Yates, S. Paquette, Emergency knowledge management and social media technologies: a case study of the 2010 Haitian earthquake, Int. J. Inf. Manag. 31 (1) (Feb. 2011) 6–13. [5] I. Guy, N. Zwerdling, I. Ronen, D. Carmel, E. Uziel, Social media recommendation based on people and tags, in: Proceeding of the 33rd International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ’10, 2010, p. 194. [6] J. Lu, D. Wu, M. Mao, W. Wang, G. Zhang, Recommender system application developments: a survey, Decis. Support Syst. 74 (2015) 12–32. [7] F. Ricci, L. Rokach, B. Shapira, Introduction to Recommender Systems handbook, in: Recommender Systems Handbook, Springer US, 2011, pp. 1–35, ISO 690. [8] M. Sohn, H. Kim, H. Lee, Personalized recommendation framework based on CBR and CSP using ontology in a ubiquitous computing environment, Int. J. Comput. Syst. Sci. Eng. 27 (6) (2012) 415–430. [9] L.O. Colombo-Mendoza, R. Valencia-García, A. Rodríguez-González, G. Alor-Hernández, J.J. Samper-Zapater, RecomMetz: a context-aware knowledgebased mobile recommender system for movie showtimes, Expert Syst. Appl. 42 (3) (2015) 1202–1222. [10] C. Rana, S.K. Jain, A study of the dynamic features of recommender systems, Artif. Intell. Rev. 43 (1) (Nov. 2012) 141–153. [11] P.D. Duffy, A. Bruns, The use of Blogs, Wikis and RSS in education: a conversation of possibilities, Queensland University of Technology, 06-Nov-2006. [12] I. Guy, N. Zwerdling, D. Carmel, I. Ronen, E. Uziel, S. Yogev, S. Ofek-Koifman, Personalized recommendation of social software items based on social relations, in: Proceedings of the Third ACM Conference on Recommender Systems, RecSys ’09, 2009, p. 53. [13] L. McAvoy, Ontological characterization and representation of context within smart environments, Comput. Syst. Sci. Eng. 30 (1) (2015) 19–32. [14] A. Stavrianou, C. Brun, Expert recommendations based on opinion mining of user-generated product reviews, Comput. Intell. (Oct. 2013). [15] R. Ghani, A. Fano, Building recommender systems using a knowledge base of product semantics, in: Proceedings of the Workshop on Recommendation and Personalization in ECommerce at the 2nd International Conference on Adaptive Hypermedia and Adaptive Web Based Systems, 2002, pp. 27–29. [16] D. Rosaci, G.M.L. Sarné, Efficient personalization of e-learning activities using a multi-device decentralized recommender system, Comput. Intell. 26 (2) (May 2010) 121–141. [17] N. Zhong, J. Liu, Y. Yao, In search of the wisdom web, Computer 35 (11) (2002) 27–31. [18] W.-P. Lee, T.-H. Yang, Personalizing information appliances: a multi-agent framework for TV programme recommendations, Expert Syst. Appl. 25 (3) (2003) 331–341. [19] N. Fornara, C. Tampitsikas, Semantic technologies for open interaction systems, Artif. Intell. Rev. 39 (1) (Jan. 2013) 63–79. [20] K. Wei, J. Huang, S. Fu, A survey of e-commerce recommender systems, in: 2007 International Conference on Service Systems and Service Management, 2007, pp. 1–5. [21] I. Garcia, L. Sebastia, E. Onaindia, On the design of individual and group recommender systems for tourism, Expert Syst. Appl. 38 (6) (2011) 7683–7692. [22] J. Golbeck, J. Hendler, Filmtrust: movie recommendations using trust in web-based social networks, in: Proceedings of the IEEE Consumer Communications and Networking Conference, University of Maryland, vol. 96, 2006, pp. 282–286. [23] Z. Hyung, K. Lee, K. Lee, Music recommendation using text analysis on song requests to radio stations, Expert Syst. Appl. 41 (5) (2013) 2608–2618. [24] Hung-Wen Tung, Von-Wun Soo, A personalized restaurant recommender agent for mobile e-service, in: IEEE International Conference on e-Technology, e-Commerce and e-Service, EEE ’04, 2004, pp. 259–262. [25] A. Martinez, J. Arias, A. Vilas, J. Garcia Duque, M. Lopez Nores, What’s on TV tonight? An efficient and effective personalized recommender system of TV programs, IEEE Trans. Consum. Electron. 55 (1) (Feb. 2009) 286–294. [26] C. Dugan, M. Muller, D.R. Millen, W. Geyer, B. Brownholtz, M. Moore, The dogear game, in: Proceedings of the 2007 International ACM Conference on Conference on Supporting Group Work, GROUP ’07, 2007, p. 387. [27] J.F. McCarthy, The virtual world gets physical: perspectives on personalization, IEEE Internet Comput. 5 (6) (2001) 48–53. [28] D. Goldberg, D. Nichols, B.M. Oki, D. Terry, Using collaborative filtering to weave an information tapestry, Commun. ACM 35 (12) (Dec. 1992) 61–70. [29] I.A. Christensen, S. Schiaffino, Entertainment recommender systems for group of users, Expert Syst. Appl. 38 (11) (2011) 14127–14135. [30] L. Martínez, M.J. Barranco, L.G. Pérez, M. Espinilla, A knowledge based recommender system with multigranular linguistic information, Int. J. Comput. Intell. Syst. 1 (3) (Aug. 2008) 225–236. [31] X. Bao, L. Bergman, R. Thompson, Stacking recommendation engines with additional meta-features, in: Proceedings of the Third ACM Conference on Recommender Systems, RecSys ’09, 2009, p. 109. [32] W.-P. Lee, Towards agent-based decision making in the electronic marketplace: interactive recommendation and automated negotiation, Expert Syst. Appl. 27 (4) (Nov. 2004) 665–679. [33] L. Lü, M. Medo, C.H. Yeung, Y.-C. Zhang, Z.-K. Zhang, T. Zhou, Recommender systems, Phys. Rep. 519 (1) (Oct. 2012) 1–49. [34] D. Mcsherry, Explanation in recommender systems, Artif. Intell. Rev. 24 (2) (Oct. 2005) 179–197. [35] C. Martínez-Costa, M. Menárguez-Tortosa, J.T. Fernández-Breis, J.A. Maldonado, A model-driven approach for representing clinical archetypes for semantic web environments, J. Biomed. Inform. 42 (1) (Feb. 2009) 150–164. [36] E. Monfil-Contreras, RESYGEN: a recommendation system generator using domain-based heuristics, Expert Syst. Appl. 40 (1) (2013) 242–256. [37] F. Ricci, Travel recommender systems, IEEE Intell. Syst. 17 (6) (2002) 55–57. [38] K. Nguyen, T. Dillon, Obligation nets: a rigorous object-oriented technique for modeling the behavioral semantics of information systems, 2007. [39] A. Rodríguez-González, AKNOBAS: a knowledge-based segmentation recommender system based on intelligent data mining techniques, Comput. Sci. Inf. Syst. 9 (2) (2012) 713–740. [40] J. Lu, Q. Shambour, Y. Xu, Q. Lin, G. Zhang, A web-based personalized business partner recommendation system using fuzzy semantic techniques, Comput. Intell. 29 (1) (Feb. 2013) 37–69. [41] A. Garcia-Crespo, Conceptual model for semantic representation of industrial manufacturing processes, Comput. Ind. 61 (7) (2010) 595–612. [42] Á. García-Crespo, J.L. López-Cuadrado, R. Colomo-Palacios, I. González-Carrasco, B. Ruiz-Mezcua, Sem-fit: a semantic based expert system to provide recommendations in the tourism domain, Expert Syst. Appl. 38 (10) (Sep. 2011) 13310–13319. [43] R. Biuk-Aghai, S. Fong, Y. Si, Design of a recommender system for mobile tourism multimedia selection, in: 2nd International Conference on Internet Multimedia Services Architecture and Applications, 2008, IMSAA 2008, 2008, pp. 1–6. [44] M. Wiesner, D. Pfeifer, Adapting recommender systems to the requirements of personal health record systems, in: Proceedings of the 1st ACM International Health Informatics Symposium, ACM, 2010, pp. 410–414.
JID:SCICO AID:1912 /FLA
18
[m3G; v1.159; Prn:16/07/2015; 14:05] P.18 (1-18)
D. Villanueva et al. / Science of Computer Programming ••• (••••) •••–•••
[45] A. Rodríguez, E. Jiménez, Semmed: applying semantic web to medical recommendation systems, in: First International Conference on Intensive Applications and Services, 2009, INTENSIVE’09, IEEE, 2009, pp. 47–52. [46] S. Gao, H. Wang, D. Xu, Y. Wang, An intelligent agent-assisted decision support system for family financial planning, Decis. Support Syst. 44 (1) (2007) 60–78. [47] P. Kazienko, K. Musial, T. Kajdanowicz, Multidimensional social network in the social recommender system, IEEE Trans. Syst. Man Cybern., Part A, Syst. Hum. 41 (4) (Jul. 2011) 746–759. [48] R. Peixoto, Traceability-based access recommendation, 2012. [49] U. Bojars, J. Breslin, S. Decker, Social networks and data portability using semantic Web technologies, 2008. [50] L. Atzori, A. Iera, G. Morabito, M. Nitti, The Social Internet of Things (SIoT) – when social networks meet the Internet of things: concept, architecture and network characterization, Comput. Netw. 56 (16) (Nov. 2012) 3594–3608. [51] Y.S. Kim, B.-J. Yum, Recommender system based on click stream data using association rule mining, Expert Syst. Appl. 38 (10) (2011) 13320–13327. [52] Y.H. Cho, J.K. Kim, S.H. Kim, A personalized recommender system based on web usage mining and decision tree induction, Expert Syst. Appl. 23 (3) (2002) 329–342. [53] K.-L. Hsieh, Employing a recommendation expert system based on mental accounting and artificial neural networks into mining business intelligence for study abroad’s P/S recommendations, Expert Syst. Appl. 38 (12) (2011) 14376–14381. [54] J. López-Cuadrado, SABUMO: towards a collaborative and semantic framework for knowledge sharing, Expert Syst. Appl. 39 (10) (2012) 8671–8680. [55] B. Towle, C. Quinn, Knowledge based recommender systems using explicit user models, in: Proceedings of the AAAI Workshop on Knowledge-Based Electronic Markets, 2000, pp. 74–77. [56] A. Tiroshi, T. Kuflik, J. Kay, B. Kummerfeld, Recommender systems and the social web, in: Advances in User Modeling, Springer, Berlin, Heidelberg, 2012, pp. 60–70. [57] H. Agius, M. Angelides, Developing knowledge-based intelligent multimedia tutoring systems using semantic content-based modelling, Artif. Intell. Rev. (1999) 55–83, no. Kanade 1996. [58] J. Glass, S. Marx, T. Schmidt, F. Sivrikaya, Semantic TV engine: an IPTV enabler for personalized recommendations, in: Inf. Manag. SPIM, vol. 34, 2010. [59] T. Di Noia, R. Mirizzi, V.C. Ostuni, D. Romito, M. Zanker, Linked open data to support content-based recommender systems, in: Proceedings of the 8th International Conference on Semantic Systems, I-SEMANTICS ’12, 2012, p. 1. [60] Y. Raimond, P. Sinclair, N. Humfrey, M. Smethurst, BBC programmes ontology, 2009. [61] I. Fernández-Tobías, I. Cantador, M. Kaminskas, F. Ricci, A generic semantic-based framework for cross-domain recommendation, in: Proceedings of the 2nd International Workshop on Information Heterogeneity and Fusion in Recommender Systems, HetRec ’11, 2011, pp. 25–32. [62] M. Ruiz-Montiel, J. Aldana-Montes, Semantically enhanced recommender systems, in: On the Move to Meaningful Internet Systems: OTM 2009 Workshops, Springer, Berlin, Heidelberg, 2009, pp. 604–609. [63] Y. Blanco-Fernández, M. López-Nores, J.J. Pazos-Arias, J. García-Duque, An improvement for semantics-based recommender systems grounded on attaching temporal information to ontologies and user profiles, Eng. Appl. Artif. Intell. 24 (8) (Dec. 2011) 1385–1397. [64] D. Tsatsou, V. Mezaris, I. Kompatsiaris, Semantic personalisation in networked media: determining the background knowledge, in: 2012 Seventh International Workshop on Semantic and Social Media Adaptation and Personalization, 2012, pp. 101–106. [65] M. Belk, P. Germanakos, N. Tsianos, Adapting generic web structures with semantic web technologies: a cognitive approach, in: Proceedings of the 4th International Workshop on Personalized Access, Profile Management, and Context Awareness in Databases, in Conjunction with VLDB, 2010, pp. 35–40. [66] T. Hofmann, Latent semantic models for collaborative filtering, ACM Trans. Inform. Syst. 22 (1) (Jan. 2004) 89–115. [67] R. Burke, A. Felfernig, M. Göker, Recommender systems: an overview, AI Mag. 32 (3) (2011) 13–18. [68] P. Crespo, C. Antunes, Predicting teamwork results from social network analysis, Expert Syst. (2013). [69] A. Catala, P. Pons, J. Jaen, J.A. Mocholi, E. Navarro, A meta-model for dataflow-based rules in smart environments: evaluating user comprehension and performance, Sci. Comput. Program. 78 (10) (Oct. 2013) 1930–1950. [70] O.C. Santos, J.G. Boticario, D. Pérez-Marín, Extending web-based educational systems with personalised support through user centred designed recommendations along the e-learning life cycle, Sci. Comput. Program. 88 (Aug. 2014) 92–109. [71] M.A. Ale, C.M. Toledo, O. Chiotti, M.R. Galli, A conceptual model and technological support for organizational knowledge management, Sci. Comput. Program. 95 (Dec. 2014) 73–92. [72] L. Chen, W. Zeng, Q. Yuan, A unified framework for recommending items, groups and friends in social media environment via mutual resource fusion, Expert Syst. Appl. 40 (8) (2013) 2889–2903. [73] M. Morzy, Cluster-based analysis and recommendation of sellers in online auctions, Comput. Syst. Sci. Eng. 22 (5) (2007) 279–287. [74] S.-M. Choi, Y.-S. Han, Representative reviewers for Internet social media, Expert Syst. Appl. 40 (4) (2013) 1274–1282. [75] M. Laroche, M.R. Habibi, M.-O. Richard, To be or not to be in social media: how brand loyalty is affected by social media?, Int. J. Inf. Manag. 33 (1) (2013) 76–82. [76] M.L. Lemos, D.V. Vasquez, M. Radzimski, L. Lemos, J.M. Gómez-berbís, RING: A context ontology for communication Channel rule-based recommender system, 2012, 1 Introduction, p. 10. [77] X.H. Wang, Da Qing Zhang, Tao Gu, H.K. Pung, Ontology based context modeling and reasoning using OWL, in: Proceedings of the Second IEEE Annual Conference on Pervasive Computing and Communications Workshops, 2004, pp. 18–22. [78] O. Santos, J. Boticario, User-centred design and educational data mining support during the recommendations elicitation process in social online learning environments, Expert Syst. (2013). [79] K. Denecke, Are SentiWordNet scores suited for multi-domain sentiment classification?, in: 2009 Fourth International Conference on Digital Information Management, 2009, pp. 1–6. [80] K. Denecke, Using SentiWordNet for multilingual sentiment analysis, in: 2008 IEEE 24th International Conference on Data Engineering Workshop, 2008, pp. 507–512. [81] M. del P. Salas-Zárate, et al., Analyzing best practices on web development frameworks: the lift approach, Sci. Comput. Program. 102 (2015) 1–19. [82] I. Wallace, M. Rovatsos, A computational framework for practical social reasoning, Comput. Intell. (Aug. 2013). [83] M. Thelwall, Sentiment in twitter events, J. Am. Soc. Inf. Sci. Technol. 62 (2) (2011) 406–418. [84] A. Tumasjan, T. Sprenger, P. Sandner, I. Welpe, Predicting elections with twitter: what 140 characters reveal about political sentiment, in: ICWSM, vol. 10, 2010, pp. 178–185. [85] G. Paltoglou, M. Thelwall, Twitter, MySpace, digg, ACM Trans. Intell. Syst. Technol. 3 (4) (Sep. 2012) 1–19.