A context-based personalized ratings management system

Abstract

In an online community of any size, every user garners reputations for different contexts through interactions with other users. When any user in the online community is interested in a certain context, the user would value the opinions of those users of high repute with respect to that context. However, if the online community is of any significant size, a user might not know every other user in the community. Therefore, a reputation brokering mechanism needs to be incorporated so that an interested user would be able to trace paths through other users to the reputable users. From social network theories, there exist centrality measures that can be used to determine the reputation of users in a network based on the number of in-degrees and out-degrees. However, different users in the network can have varying tastes and opinions with respect to a given context. Since centrality measures determine the reputation of users based on aggregate opinion, a user who has different tastes from the majority of the other users might not agree with the reputable users selected by the centrality measures. This justifies the need for developing personalized rating systems that are able to personalize for any user a selection of other users that he would regard as highly reputable. In this thesis, two such rating systems are developed and compared against the existing centrality measures. When tested over various dimensions such as network size and network connectivity, there is evidence that the personalized rating systems perform better than the traditional measures of reputation in the selection of reputable individuals.