Preference and Sentiment Guided Social Recommendations with Temporal Dynamics

Abstract

Capturing users' preference that change over time is a great challenge in recommendation systems. What makes a product feature interesting now may become the accepted standard in the future. Social recommender systems that harness knowledge from user expertise and interactions to provide recommendation have great potential in capturing such trending information. In this paper, we model our recommender system using sentiment rich user generated product reviews and temporal information. Specifically we integrate these two resources to formalise a novel aspect-based sentiment ranking that captures temporal distribution of aspect sentiments and so the preferences of the users over time. We demonstrate the utility of our proposed model by conducting a comparative analysis on data extracted from Amazon.com and Cnet. We show that considering the temporal preferences of users leads to better recommendation and that user preferences change over time.