Leveraging Community Detection for Accurate Trust Prediction Ghazaleh Beigi Department of Computer Engineering Sharif University of Technology ghbeigi@ce.sharif.edu Mahdi Jalili Department of Computer Engineering Sharif University of Technology mjalili@sharif.edu Hamidreza Alvari Department of EECS University of Central Florida halvari@eecs.ucf.edu Gita Sukthankar Department of EECS University of Central Florida gitars@eecs.ucf.edu ABSTRACT The aim of trust prediction is to infer trust values for pairs of users when the relationship between them is unknown. The unprecedented growth in the amount of online interactions on e-commerce websites has made the problem of predicting user trust relationships crit- ically important, yet sparsity in the amount of known (labeled) relationships poses a significant challenge to the usage of machine learning techniques. This pa- per presents a community detection approach which leverages the network of available trust relations and rating similarities to compensate for the lack of labels. The key insight behind our framework is that trust values from the central community members can be used as a predictor for relationships between other community members. Here we evaluate the usage of two community detection algorithms, one of which works merely on the trust network while the other one uses both. Our algorithm outperforms other existing trust prediction methods on datasets from the well- known product review websites Epinions and Ciao. I INTRODUCTION Trust prediction, the ability to identify how much to trust to allocate an unknown user, is an important prerequisite toward the development of scalable on- line e-commerce communities. We are more likely to purchase an item from a seller on an e-commerce web- sites such as eBay or Amazon, if our trusted acquain- tances have reported positive experiences with that seller in the past. Reviews from trusted users will carry more weight towards the purchasing decision than reviews from anonymous or unknown customers. Trust can be gained or lost through direct personal interactions, but this is impractical for popular e- commerce systems which boast millions of users. Thus, these platforms must support computational mech- anisms for propagating trust between users. This problem is complicated by the fact that most cus- tomers only have interactions with a small set of other users and products, resulting in a sparse dataset of known trust relationships. In this paper, we propose a novel community-based mechanism for propagating trust between users, even when they are not closely connected by existing links. The underlying assumption is that customer trust values are likely to be strongly correlated with other customers within the same community. Using com- munity detection, users are grouped into non-exclusive communities (i.e., each user can be a member of sev- eral communities), which are represented by a pro- totypical highly-connected community member. Our model uses the community membership vector to in- fer trust values between two users by examining the similarities between the users and representative com- munity members. This paper introduces a two-phase approach to pre- dict the trust values between each pair of users. In the first phrase, we cluster users into communities. This paper evaluates the usage of two different commu- nity detection algorithms: a game-theoretic approach (originally introduced in [1]) that operates under the assumption that users join communities to maximize their utility, which is calculated from a combination of rating similarity and the network neighborhood of known trust relations. For the second algorithm we use smart local moving (SLM) community detec- tion [2] which detects communities by maximizing a modularity function. SLM is only designed to work on a single network, so we we run it on trust network only. In the second phase, we predict the trust between each pair of users by comparing the similarities be- tween their respective community membership vec- tors. To calculate the similarity between two commu- nities, our community-based algorithm compares the central, or prototypical, community members. This paper evaluates the relative merits of different cen- 2014 ASE BIGDATA/SOCIALCOM/CYBERSECURITY Conference, Stanford University, May 27-31, 2014 ©ASE 2014 ISBN: 978-1-62561-000-3 1