Boosting Collaborative Filtering by Trust

Abstract

As recommendation systems continue to play a major role in users' daily interactions on the web, they continue to be heavily researched. The emergence and widespread use of social networks and the accessibility of personal information have made it possible to build recommendation systems that exploit novel concepts such as trust and/or the personal social network structure. Collaborative filtering is one of the most commonly used algorithms in recommendation systems. This research leverages the abilities of collaborative filtering by introducing trust-based and social factors to modify existing similarity and neighborhood measures. Using this hybrid approach, a 45% reduction in error rate has been shown as measured by root mean squared error, comparing baseline collaborative filtering with trust-aware collaborative filtering. The experiments in this research have been conducted using multiple variables in collaborative filtering including neighborhood size, and threshold values in neighborhood formation sub-process. Moreover, experiments were also conducted on multiple similarity measures to identify the best combination of neighborhood formation and similarity that gives an optimal solution in a social trust context.