IADIS International Journal on WWW/Internet Vol. 7, No. 1, pp. 13-25 ISSN: 1645-7641 13 INFERRING TRUST BASED ON HIGHLY TRUSTED USERS IN A RECOMMENDER SYSTEM Yussuf Abu-Shaaban. Department of Computer Science, School of Mathematical and Computer Sciences, Heriot Watt University, Edinburgh, EH14 4AS, UK. E-mail: ya37@hw.ac.uk ABSTRACT Trust information is becoming essential to support users making decisions when interacting e-commerce services and social networks. Such trust information is usually represented by a directed network of nodes (users) connected by trust edges. Trust propagation and aggregation techniques are usually used to infer the trust value between two users with no direct trust link. In this paper, we propose and investigate a new approach to infer trust by using recommender system information. To determine the trust value of a user, ratings assigned by the highest trusted users to recommendations issued by that user are aggregated to infer trust. It was found that basing trust decisions on recommendation ratings made by the highest trusted users can achieve high accuracy outperforming less trusted users. In addition, trust value inference based on recommendation ratings of the highest trusted users tend to give a trust value greater than the mean of the direct trust values in a network of trust. This tendency can also be noticed, with a smaller degree, when trust is predicted based on recommendation ratings made by less trusted users. The work presented in this paper is the first to use recommender systems to predict trust and to analyse the accuracy and tendencies of their use in the trust inference process. KEYWORDS Trust, Trust propagation, Trust Inference, Recommender Systems. 1. INTRODUCTION With the exponential growth of the Internet and its services, techniques to aid users to make decisions on the trustworthiness of services and other users are becoming more vital (Klang 2001). With the increasing importance of trust information in e-commerce systems and social networks, there has been more interest in the way such trust information is represented and inferred. A web of trust is usually constructed, such as Eopinions.com, where users express their trust in other users to form a directed graph where nodes represent users and trust values