Graph-based opinion entity ranking in customer reviews

Online product reviews currently pose large impact on purchasing decision of potential customers. However, the overwhelming number of those reviews hinder people to find useful information and make good decisions on the purchases. In this paper, we propose a graph-based opinion entity ranking framework to mine opinion data from former customers, and rank either entities (products) or aspects (features) in accordant with those opinions. From the customer reviews, we first extract aspects and their sentiment words for each entity. We then represent relationships between reviewers and pairs of entity-aspect by a weighted bipartite graph, and propose an algorithm to compute the ranking scores. Experimental results on a hotel review dataset show higher ranking agreements with those of the human users than ones from tradition frequency-based baseline.