Experimental Evaluation of Deceptive Information Filtering in Context-Aware Service Selection ⋆ Murat S¸ensoy and Pınar Yolum Department of Computer Engineering, Bo˘ gazic¸i University, Bebek, 34342, Istanbul, Turkey {murat.sensoy,pinar.yolum}@boun.edu.tr Abstract. Experience-based service selection refers to selection of service providers using others’ experiences. An agent can represent its experience (its demand and received service) rigorously using an ontology. When agents report their experiences truthfully, experience-based service selection outperforms clas- sical rating-based service selection, in which agents only report ratings for service providers. However, in many setting agents may prefer to lie about their experi- ences. This paper tackles the problem of handling deceptive information in the context of experience-based service selection. We apply three current approaches for filtering unfair ratings to filtering deceptive experience. We analyze these ap- proaches when multiagent systems have different types of liars and report their performance in filtering deceptive experiences. 1 Introduction Service selection refers to the problem of finding service providers for a given task in a particular context. Traditional approaches to service selection are based on the ex- change of rating among agents in a multiagent system. That is, typically in rating-based approaches, the consumers rate the service providers and share these ratings with other consumers. Then, the shared ratings are aggregated to figure out the most satisfactory service providers. However, just by looking at the rating, it is difficult if not impossi- ble to understand in what context and under what expectations a rating has been given. More importantly, without proper understanding of these, it is difficult to use ratings as a basis for further reasoning on service providers. To deal with these shortcomings, we have previously proposed to use experiences as opposed to ratings to represent past dealings with providers [1]. An experience roughly corresponds to a user reviews on Web sites in that it captures an episode of the customer with a provider and can be thought of as a record of what service the customer has requested and received in return. Contrary to user reviews, an experience contains only objective information rather than subjective evaluations. For instance, while a user review in real-life may state that the price of the experienced service was fair, an experience states only the actual price of the service (e.g., $100). Further, an experience is represented using a common ontology, which makes it possible for the agents to understand and reason on the experiences. ⋆ This research has been partially supported by Bo˘ gazic¸i University Research Fund under grant BAP07A102 and The Scientific and Technological Research Council of Turkey by a CAREER Award under grant 105E073. R. Falcone et al. (Eds.): TRUST 2008, LNAI 5396, pp. 326–347, 2008. c  Springer-Verlag Berlin Heidelberg 2008