Research On Opinion Analysis For Book Reviews

With the increasing popularity of network and computers, customers can express their reviews after shopping online. As these reviews contains consumers'attitudes of products. So, mining product reviews plays a supporting role at improving the product for trade companies and purcharing the product for consumers, which is valuable in application and research.Product reviews mining includes the research of features identification, opinion words extraction and polarity classification. In these researches, previous methods all focused on the content of reviews, and the polarity dictionary constructed is to be composed of fixed words, which affects the effective of reviews mining. To solve these problems, we devote the research on book reviews. The main tasks of this paper are as follows:Firstly, we consider that the contribution of words to categories is different, so use CHI to construct dictionary and propose a polarity dictionary construction method based on the improved CHI, which classifies words via calculating the CHI of each word. Then, this paper extract the words which are not included in the dictionary according to the characteristics of co-occurrence of similar words, which realizes dynamic addition of the dictionary and solves the problem of fixed dictionary to some extent. Considering some polar word can only modify a certain feature, so we furtherly classifies the polar words to be used for summaring reviews. In analyzing the polarity of reviews, the polarity calculation formula of the transitional complex sentences is improved to be applicable to book reviews. Besides, considering that some book reviews have titles and these titles generally express the reviewers'opinion, so we propose an opinion polarity analysis method based on the titles and the improved polarity calculation formula of the heavy transitional sentences. This method considers the polarity of titles as the polarity mark of reviews for analyzing the reviews polarity and uses the improved formula adjusts the reviews polarity, which reduces the errors.In summaring the reviews, we improve the SBV algorithm to be applicabble to book reviews. This method extracts features and opinion words according to dependency relation of words, and then summary the reviews.Experimental results show that the approach proposed in this paper is effective and effectively improved the effect of opinion annalysis for book reviews.