Review Summarization Based on Linguistic Knowledge

Abstract

In this paper, domain-knowledge extraction and aspect- opinion extraction are proposed in order to generate a summary from the relevant product and service review. In order to extract the word corresponding to aspect and opinion, we extract the domain-salient word and collocation information by applying statistical techniques from the bulk of the text, and construct the clue words through manual filtering. In domain knowledge extraction, in order to extract useful information, domain-salient words which occur more significantly in a given domain rather than in a public domain article are automatically extracted by using the statistical techniques. As well, collocation information has the association with high frequency words. In recognition of aspect-opinion association, words corresponding to aspects and opinions in a sentence are checked by using information of clue words, and the polarity of the sentence is determined by performing pattern-based modality analysis. Through checking the binary association based on the frequency of co-occurrence, a pair of aspect and opinion is extracted, our system can automatically acquire the scores for a review target based of the degree of positive/negative.