Summarizing customer reviews through aspects and contexts

Abstract

This study leverages the syntactic, semantic and contextual featuresof online hotel and restaurant reviews to extract information aspects and summarizethem into meaningful feature groups. We have designed a set of syntacticrules to extract aspects and their descriptors. Further, we test the precision ofa modified algorithm for clustering aspects into closely related feature groups,on a dataset provided by Yelp.com. Our method uses a combination of semanticsimilarity methods- distributional similarity, co-occurrence and knowledge basebased similarity, and performs better than two state-of-the-art approaches. It isshown that opinion words and the context provided by them can prove to begood features for measuring the semantic similarity and relationship of theirproduct features. Our approach successfully generates thematic aspect groupsabout food quality, décor and service quality.