Review Extract Using Word Embeddings

Abstract

Online reviews become the basis of evaluating alternatives before making purchase decision. So extracting useful information from a large number of reviews becomes a valuable act. Recently word embeddings methods have been widely used in binary sentiment classification tasks and perform well. They judge a review is negative or positive, but ignores the product characteristics of the reviews expression. It’s difficult to get valuable information from huge number of reviews. In this paper, we propose a method to get the core reviews from review collection. This methodology applies a combined approach of word embedding (word2vec) and PageRank named WVP to calculate the impact factor of a review. We training the model through an open source hotel review data. The methodology is applied to the data sets of hotel reviews was crawled from DianPing which a famous company on tourism. Experiment show that WVP is effective to get core reviews.