Sentiment and preference guided social recommendation

Abstract

Social recommender systems harness knowledge from social experiences, expertise and interactions. In this paper we focus on two such knowledge sources: sentiment-rich user generated reviews; and preferences from purchase summary statistics. We formalise the integration of these knowledge sources by mixing a novel aspect-based sentiment ranking with a preference ranking. We demonstrate the utility of our proposed formalism by conducting a comparative analysis on data extracted from Amazon.com. In particular we show that the performance of the proposed aspect based sentiment analysis algorithm is superior to existing aspect extraction algorithms and that combining this with preference knowledge leads to better recommendations.