Fuzzy computational models for trust and reputation systems
Introduction
The emerging online environment let people buy and sell goods, play, and recommend products for each other without knowing the remote partner. All of these encounters are only possible if partners trust each other. Consequently they can delegate tasks and decisions to an appropriate person and therefore improving the quality of online markets [1], [2]. However, insufficient trust in e-commerce could lead to users ‘staying away’ from the technology altogether. Despite obvious importance of trust and reputation, transferring them to computer science is still a challenging problem. Computer scientists from many areas, e.g., security, semantic web, and e-commerce, are still working on the transfer of these concepts to be used in computer mediated transactions. It is often hard to assess the trustworthiness of remote partners in the computer mediated transactions and processes because familiar styles of interaction are still far from being computationally modeled easily. Actually, physical transactions and traditional forms of communication like body language, gestures, and facial expressions allow people to assess a much wider range of cues related to trustworthiness than the current computer mediated communication can do.
Amongst different areas of computer science, e-commerce seems to be the most benefitted area from the trust and reputation concepts. Recently, very big electronic communities have been established and grown very fast. Unfortunately, a major weakness of e-markets is the raised level of risk associated with the loss of the notions of trust and reputation. The consumer is usually forced to accept the ‘risk of prior performance’, i.e. to pay for services and goods before receiving them, which can leave him in a vulnerable position. On the other hand, the service provider has much more information about the product quality than the customer, as long as he will receive payment before shipment in most cases. The research work in the recent past [1], [3], [4], [5], [6], [7], [8] shows that this information asymmetry can be mitigated through trust and reputation concepts. The idea is that even if the consumer cannot try the service in advance, he can be confident that it will be what he expects as long as he trusts the service provider.
Various definitions have been given in the literature to both trust and reputation concepts. For example, Ruohomaa and Kutvonen [7] defined trust as the extent to which one partner is willing to participate in a given action with another partner, considering the risks and incentives involved. On the other hand, the reputation is defined by Mui et al. [8] as a perception a partner creates through past actions about his intentions and norms. Although the way of incorporating trust and reputation concepts in e-commerce varies from system to system but they provide an incentive for good behavior and a punishment for bad behavior. However, trust and reputation concepts are social fuzzy concepts thus fuzzy computational models for both the concepts would reflect their actual value for enhancing the system accuracy.
The rest of this paper is organized as follow: trust and reputation concepts in literature are given in Section 2. The proposed fuzzy computational models for trust and reputation are presented in Section 3 and are employed for movie RS in Section 4. Computational experiments and results are given in Section 5. Finally, the last section concludes the paper and gives some future research directions.
Section snippets
Trust and reputation in literature
Trust and reputation concepts get much attention due to the need for reliable automatic tools for Web services. The traditional cues of trust and reputation of physical encounters are missing in Web interactions. As a consequence, electronic substitutes are required for these concepts so that the quality of online services can improve especially for e-commerce and entertainment services. However, trust and reputation are quite challenging to define as they manifest themselves in many different
The proposed fuzzy computational models for trust and reputation concepts
The basic idea of trust and reputation systems is to let partners rate each other after the completion of an encounter. The system then aggregates these ratings for a given partner to derive a personalized trust measure or a global reputation score [1]. However, either trust or reputation are usually used as crisp attributes. That does not effectively reflect the social meanings of these two concepts where most of human perceptions are fuzzy. Our attempt in the following section is to discuss
Trust and reputation models for movie recommender system
Recently, RS have become an essential part of e-commerce online services where they offer suggestions about products customers might also like to buy. Their contribution comes in two forms, either to predict ratings of products that a user wants to know about, or to list products that users might find of interest based on their stated preferences, online shopping choices, and the purchases of people with similar tastes or demographics. This would create new revenue opportunities and increasing
Experiments
The MovieLens dataset consists of 100,000 ratings, assigned by 943 users on 1682 movies. All ratings follow the 1-bad, 2-average, 3-good, 4-very good, and 5-excellent numerical scale. Each user has rated at least 20 movies. In our experiments, we considered only users who have rated at least 59 movies, 20 (34%) for user profiling and 39 (66%) for testing. Out of 943 users, 500 users satisfied this condition and contributed 84,773 ratings out of 100,000. This dataset is used as the basis for
Conclusion
General computational models are proposed for reputation and trust in this paper. The proposed models can be implemented for any application as long as it is a rating system explicitly or implicitly. We have used our models in building movie RS to enhance the recommendation accuracy. Incorporating trust and reputation concepts separately gives a two-level filtering methodology to enhance the recommendation accuracy through reputation-based similarity and trust-based filtering. A comparison of
Acknowledgment
The authors would like to thank the anonymous reviewers for their valuable comments.
References (33)
- et al.
A survey of trust and reputation systems for online service provision
Decision Support Systems
(2007) A computational approach to fuzzy quantifiers in natural languages
Computer and Mathematics with Applications
(1983)- et al.
Fuzzy-genetic approach to recommender systems based on a novel hybrid user model
Expert Systems with Applications
(2008) - S. Ries, J. Kangasharju, M. Muhlhauser, A classification of trust systems, in: OTM Workshops, LNCS 4277, 2006, pp....
- et al.
Review on computational trust and reputation models
Artificial Intelligence Review
(2005) - J. O’Donovan, B. Smyth, Trust in recommender systems, in: Proceedings of IUI’05, San Diego, CA, USA, 2005, pp....
- M. AL-Mutairi, Preference uncertainty and trust in decision making, Ph.D. Thesis, University of Waterloo, Waterloo,...
- et al.
Collaborative filtering using dual information sources
IEEE Intelligent Systems
(2007) - S. Ruohomaa, L. Kutvonen, Trust management survey, in: P. Herrmann, et al. (Eds.), iTrust 2005, LNCS 3477,...
- L. Mui, M. Mohtashemi, A. Halberstadt, A computational model of trust and reputation, in: Proceedings of the 35th...