trust in electronic commerce - Semantic Scholar

Expert Systems with Applications 28 (2005) 623–628 www.elsevier.com/locate/eswa

A fuzzy logic-based system for assessing the level of business-to-consumer (B2C) trust in electronic commerce Fahim Akhtera,*, Dave Hobbsb, Zakaria Maamarc a

College of Information Systems, Zayed University, Box 19282, Dubai, UAE b University of Bradford, Bradford, West, Yorkshire, UK c Zayed University, Dubai, UAE

Abstract The purpose of this paper is to present an application of fuzzy logic to human reasoning about electronic commerce (e-commerce) transactions. This paper uncovers some of the hidden relationships between critical factors such as security, familiarity, design, and competitiveness. We analyze the effect of these factors on human decision process and how they affect the Business-to-Consumer (B2C) outcome when they are used collectively. This research provides a toolset for B2C vendors to access and evaluate a user’s transaction decision process, and also an assisted reasoning tool for the online user. q 2005 Elsevier Ltd. All rights reserved. Keywords: Business-to-Consumer; e-commerce; Fuzzy logic

1. Introduction and motivation During online shopping, a user often relies on common sense and applies vague and ambiguous terms when making a buying decision. Online customer normally develops in his/her mind some sort of ambiguity, given the choice of similar alternative products and services (Mohanty & Bhasker, 2005). Decisions to buy or not to buy online are often based on users’ human intuitions common sense and experience, rather than on the availability of clear, concise and accurate data. Fuzzy logic is used for reasoning about inherently vague concepts (Lukasiewicz, 1970), such as ‘online shopping is convenient’, where level of convenience is open to interpretation. The purpose of this research is therefore to apply the fuzzy logic to human reasoning where we specifically focus on the reasoning processes behind e-commerce transactions. Fuzzy systems allow the encoding of knowledge in a form that can be used to reflect the way humans think about a complex problem such as online shopping. A human usually think in imprecise terms such as high and low, fast * Corresponding author. Tel.: C97150 6743130; fax: C9714v2640854. E-mail addresses: [email protected] (F. Akhter), d.hobbs@ bradford.ac.uk (D. Hobbs), [email protected] (Z. Maamar). 0957-4174/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2004.12.039

and slow, and heavy and light (Black, 1937). Fuzzy expert system model imprecise information, by attempting to capture knowledge in a similar fashion to the way in which it is considered to be represented in the human mind, and therefore improves cognitive modelling of a problem (Cox, 1994). As a result, fuzzy logic is leading to new and humanlike, intelligent systems that might be used to understand the thought processes behind any B2C transactions. The rationale for using fuzzy logic systems to uncover vague decision process because it is well suited for modeling human decision-making. Human decision-making is complex, and can be based on simultaneous evaluation of many facets such as fear, experience, privacy, intuition and so forth. Though many factors influence the decision process of B2C transactions such as ease-of-use, pricing, convenience, and security (Akhter et al., 2003), the perception of an influencing feature is more important than the actual level of the feature itself. For example if the perceived security level is higher than its actual implementation then that will contribute positively to the level of B2C outcome. There may be cases where the reverse is true as well, but for such cases a high level of persuasion will be needed to alter the perception level. This research had adopted a fuzzy logic approach and utilized a mathematical research toolset

624

F. Akhter et al. / Expert Systems with Applications 28 (2005) 623–628

known as Matlab fuzzy logic toolboxw to provide a means of coping with the ambiguity and vagueness that are often present in determining a transaction level in e-commerce. To build a fuzzy expert system for B2C e-commerce that is based on fuzzy logic, this research has captured, organised and used human expert knowledge (acquired by surveys and interviews). This research proposed to organise knowledge in terms of its logical groupings such as security, familiarity, design layout, competitiveness and trust. This paper is organized as follows. Section 2 discusses research methodology that demonstrates the value of qualitative technique for inquiry and analysis of data. Data collection and analysis explained in Section 3. Section 4 covered the rules providing a measure for Trust and B2C levels. Section 5 explained the analysis of different factors influencing Trust. Section 6 visualizes the Trust level and Section 7 draws our conclusions.

2. Methodology This research is based on the rationale that actual level of any B2C transaction is based on two factors, namely: what is the level of Trust (T) of the given website, and how competitive (C) is this site for purchasing purposes? Therefore, we propose to investigate into the truthfulness of the following relationships: T Z f ðS; F; DÞ

(1)

LB2C Z gðT; CÞ

(2)

Where S is the level of security, F is the level of familiarity, and D is the level of design layout of the B2C site. The premise is that the factors determining the level of Trust T are a function of these three parameters. Therefore any degree of B2C transaction will be based of the level of Trust (T) and the competitiveness (C) of the website. In order to analyse the impact of human decisions on the level of e-commerce transactions, it was organise and categorise the factors that are significant to the decision processes linked to conducting the B2C transaction. It is apparent from Fig. 1 that a given level of Trust may lead to a membership of more than one fuzzy set. This membership is represented by its degree or intensity. Therefore, as a consequence this may lead to the partial execution of the antecedent (one or more premises) and subsequent partial execution of the consequent of the fuzzy rules. The total numbers of rules depend on the number of hedges for each fuzzy set. Trust will have five fuzzy subsets, each of which are given by their membership function as depicted in Fig. 1. Similarly the security, familiarity and design can be divided into three levels (low, moderate and high). Hence the number of fuzzy rules for determining the level of trust can be derived as: security (three), familiarity (three) and design layout (three), which combined results in 27 distinct fuzzy rules. In order to get a complete picture of fuzzy expert system, an inference diagram can give a detailed explanation of the processes involved. The picture in Fig. 2 attempts to summarise the steps and processes involved. As can be seen, the process with the crisp inputs to the fuzzy expert system; for example, this might be the crisp input for security, and familiarity or design to get a value for the trust level.

Fig. 1. Displays the membership functions for different fuzzy sets belonging to trust.


625

Fig. 2. Complete fuzzy expert system.

Similarly a crisp level of Trust and Competitiveness will be required as inputs to the second level inference as given in Eqs. (1) and (2). It should be noted that the initial input(s) are a crisp set of numbers. These values are converted from a numerical level to a linguistic level. Following that the fuzzy rules are applied and fuzzy inference engine is executed. This will result in a given B2C level as varying degree of membership of fuzzy subsets of the B2C superset. The last step that is the defuzzification process, at which time we extract a numeric value for likelihood of the B2C transaction.

3. Data collection and analysis This study used a web-based survey because of its advantages such as convenience, viable, effective way to access difficult-to-reach respondents. On the other hand, a web survey has some limitation such as unequal opportunity (David, 2000). Even though only respondents who have access to the Internet are able to participate in this survey, this condition is exactly what is desired for respondents of this study, which is why a web-based survey was chosen. A pilot test was conducted to test the instrument prior to collecting data. The purpose of the pilot test was to access whether the instruments were capturing the phenomena desired. The Website published on the Intranet and the URL

was provided to the subjects who were enrolled in information systems and computer science classes. Respondents were asked to analyse three websites: Amazon.com, Lastminute.com and Uaemall.com for an item of their choice using a credit card. Respondents were asked to go through the entire buying process at the three Websites but told not to click the buy button to purchase the item. During this process, they had to respond to the questions under five categories namely: security, familiarity, designs, trust, competitiveness, and finally to choose the appropriate B2C level. There were four questions on security, familiarity, and design, one on trust, four on competitiveness and one on B2C level. After the Website analysis and answering the questions, lastly subjects had to identify the trust and the B2C level of the specific Website. Respondents were next asked to rank the trust level of the Website which they had finished analysing. They had a choice of ranking the Trust level as follows: Very low Trust, Low Trust, Moderate Trust, High Trust and Very high Trust. The chosen values were picked as per their understanding of Trust following the questions pertaining to Security, Familiarity and Design. The order was deliberately set so that the user could make an ‘informed’ decision based on the previous answers. At the end of the survey, respondents were asked to rate the chosen Website for B2C transactions based on their perceived level of trust and competitiveness. This level of

626


B2C will be compared by the level produce by fuzzy logic later on in the study. Respondents has been asked to rank the Trust level of the website which they have finished analysing. This B2C level would depend on their understanding of the Trust which they had already taken a stance on and the Competitiveness of the website.

4. Rules indicating trust and B2C levels Table 1 displays the user’s preferences for B2C level based on their perceived level of trust and competitiveness. The rules describing the basis for a given Trust level was based on degrees of security, familiarity and design. These degrees were formulated in terms of their linguistic variables such as low, moderate and high. The degree for Design level was expressed in terms of poor, moderate and good. Similarly the degree for a Trust level was ranging from very low to very high, in five distinct fuzzy sets. These rules were derived from the survey data after a thorough organization and analysis and represent the users views of the Trust level of a given website based on the given factors. A rule from Table 2 above can be extracted as: If ðsecurity Z highÞ and ðfamiliarity Z lowÞ and ðdesign Z moderateÞ then ðtrust Z moderate trustÞ: Trust is given in terms of five fuzzy sets whilst the competitiveness and B2C level is represented in terms of three linguistic labels for fuzzy sets. A given rule from Table 1 can be expressed as: If ðtrust Z low trustÞ and ðcompetitiveness Z highlyÞ then ðB2C level Z moderateÞ

Table 2 Formation of Trust rules Rule no.

Security linguistic value

Familiarity linguistic value

Design linguistic value

Trust linguistic value

1 2 3 4 / / 23 24 25 26 27

High High High High / / Moderate Moderate Moderate Moderate Moderate

High High High Low / / Low Low Moderate Moderate Moderate

Good Moderate Poor Good / / Moderate Poor Good Moderate Poor

VHT VHT HT HT / / LT LT MT LT LT

VHT, very high trust; HT, high trust; MT, moderate trust; LT, low trust.

5. Analysis of Trust versus security factor In order to fully understand the contributions from various factors contributing to the Trust level it is required that we examine contribution from each factor separately. The Fig. 3 shows contribution to Trust of a given Website originating from the Security. Therefore, the contribution from Familiarity and Design has been kept constant at three levels, namely: low, moderate and high corresponding to numeric values for Familiarity and Design of (1–4 and 7). Fig. 4 shows that Trust level is monotonically increasing for increasing perceived security of a website for any given level of Familiarity and Design (F and D). However when both F and D is ‘High’ (numeric value of 7) the Trust level is at its maximum for maximum Security. The three curves have one common feature that they exhibit a ‘staircase shaped’ curvature.

There were a total of 27 rules for Trust deduced from the survey as shown in Fig. 3. Similarly the rules disclosing the B2C level for various inputs were found to be 15. These rules form the basis of the compounded inference system consisting of two separate, but interconnected systems to postulate a B2C level for given values for security, familiarity, design and competition levels of the website. Table 1 Formation of B2C rules Rule no.

Trust linguistic value

Competitiveness linguistic value

B2C level linguistic value

1 2 3 / / 13 14 15

HT HT HT / / VLT VLT VLT

Fairly Highly Moderate / / Fairly Highly Moderate

Low High Moderate / / Low Low Low

VLT, very low trust; HT, high trust; MT, moderate trust; LT, low trust.

Fig. 3. Shows the GUI for the rule editor.


627

‘moderate’ and Security is ‘moderate’ to ‘high’. This suggest that when people are somewhat familiar with a website then a small increase in security levels from between moderate to high security will boost their trust in a significant way. Looking at Fig. 5 diagonally from (low, low) to (high, high) levels of Security and Familiarity one observes three plateaus where the last one is around 0.925, and remains at that level even when the input factors are increased further. This result is somehow unexpected and may be due to the fuzzy nature of the expert system where a ‘Trust’ or ‘Truth’ level of 100% is unrealistic.

7. Conclusion

Fig. 4. Trust versus Security for constant familiarity and design.

It is interesting to note that for ‘low’ and ‘moderate’ levels of F and D the developed Trust is almost identical up to a Security level of about 5. Then there is a sharp change on the Trust level between ‘low’ and ‘moderate’, and the perceived Trust for ‘moderate’ F and D is approaching that of ‘high’. A general observation is that Trust is positively related to Security for any given value of Familiarity and Design. This observation is also plausible to the Human mind. One feature that is disclosed from this figure is that for ‘high’ levels of Security the Trust difference is less significant for ‘moderate’ and ‘high’ levels of F and D. This result could not be anticipated from the outset.

6. Visualization of trust as function of security and familiarity We now attempt to visualize the Trust level as a continuous function of its input parameters. Fig. 5 attempts to portray variation of Trust as encapsulated in the rules for Trust. The highest gradient for Trust is when Familiarity is

Fig. 5. Trust level is positively related to levels of security and familiarity.

The e-commerce has given increased choices to consumers due to the growth in the number of online Websites offering products with many variations. In our paper, a tool is defined to assist consumers and vendors to analyse the level of perceived trust in a specific Website. The consumers can broadly be categorized into two groups, namely those who are technically critical of a site and capable of measuring its security features and those who are not. This survey can be used to step by step follow the instructions and based on actual level of a feature decide its contribution in a category and consequently derive a total value of a factor say Security. Hence the survey can make a buying decision more solid, based on actual appearances of various features. An added advantage would be to feed this data to the FIS for Trust and B2C and the user could compare his/her buying decision with that of others based on the outcome of the fuzzy expert system. For those who are not necessary technically inclined this survey will assist them in trying to gauge the presence of a feature, say security seals and attach to it a certain contribution (i.e. yesZ2 noZ0 and don’t knowZ1). After all the requirements for Trust is completed the FIS for Trust could be used to provide a perceived level that the user could compare with his own. Similarly the B2C level could both be guessed by the non-technical user and also computed based on input valued and presented to the user for comparisons. Now since this category of users is nontechnical the perceived B2C level could also be compared with previous users who were technically inclined and presented as a possible different B2C level to reflect upon. This procedure would assist even the non-technical user to make an informed online purchase. The vendor would benefit from the survey data that is aggregated over time and is used to amend or refine existing rule-sets. Since the data would be accumulated over time the responses would be a blend from both technical and nontechnical users. Hence the actual occurrence of a feature would be replaced by its perceived equivalence. Since the existence of a feature is only relevant to the user if it can be acknowledged, and if it can not then the vendor must seriously reconsider inclusion of this aspect on the website.

628


In addition the vendor can use the survey data to ascertain the Trust level of the site as per user’s perception and rectify if needed if this is not obvious or is having a negative impact on the Trust level. Furthermore a measure of the competitiveness is directly deductible from this survey and could be used to retain or increase market share. Lastly as the usage of the survey procedure matures (possibly by providing incentives as discounts on a completed transaction) the Fuzzy Inference Systems could be modified and adjusted where necessary. One limitation of the constructed FIS of this study is that all premises in the antecedent part of a rule have been connected with AND operation where OR operations could also be deployed. The implication and aggregation from the rule would then be significantly different.

References Akhter, F., Hobbs, D., & Maamar, Z. (2003). How users perceive trust in virtual environment. International Conference on Information and knowledge engineering, Las Vegas, Nevada, 23–26. Black, M. (1937). Vagueness: An exercise in logical analysis. Philosophy of Science., 4, 427–455. Cox, E. (1994). The fuzzy systems handbook A practitioner’s guide to building, using, and maintaining fuzzy systems. Cambridge: Academic Press. David, J. L. (2000). You’ve got Surveys. American Demographics , 42–44. Lukasiewicz, J. (1970). Philosophical remarks on many-valued systems of prepositional logic Reprinted in Selected Works. Amsterdam: NorthHolland. pp. 153–179. Mohanty, B. K., & Bhasker, B. (2005). Product classification in the Internet business—a fuzzy approach. Journal of Decision Support Systems, 38, 611–619.