Research On The Approaches Of Mining Product Features From Chinese Customer Reviews On The Internet

With the deeper and wider applications of the Internet in past ten years, more and more customers browse large number of online reviews in order to know other customers word-of-mouth of product and service to make an informed choice. At the same time, the network customer reviews as a feedback mechanism can help vendors and manufacturers improve their products and service, and then get competitive advantage. However, with the e-business arising, the number of reviews is growing rapidly and the content is more complicated. As a result, it is very difficult to retrieve useful knowledge from customers'reviews. It needs technical methods to improve the accuracy and convenience of mining information. Up to now, the technique is still a complicated task with great challenge. Review mining is to extract valuable information from customers'reviews, and it has attracted many researchers'attention. As a new area of unstructured information mining it mainly includes sentiment classification, mining products features and learning subjective language etc. In English reviews area, Researchers have made some successful results but few studies have been conducted to Chinese customer reviews on the internet. As Chinese e-business has increased dramatically in cyber space, how to automatically retrieve useful knowledge from online Chinese reviews has become urgent. However,because of the differences in characteristic between the two languages of Chinese and English, existing English oriented approaches were hard to imply directly on Chinese. This work just focus on Chinese customer reviews on the internet, and explore the technology of product features mining, in order to provide a more convenient and scientific tool for enterprises and customers in Chinese e-commerce field.Firstly, this thesis consumes the network customer reviews as the word of mouth online, and a thorough structural theory analysis of Chinese customer reviews mining is performed. On top of the problem about mining product features from Chinese Customers Reviews, the thesis constructs the DFM (Data-Function-Method) model; it also proposes a research approaches framework of mining product features from Chinese customer reviews on the internet.For extracting the product features which customers are concerned, this study develops the principles of English review mining methods, and based on the theory of association rules, in particular, Apriori algorithm extract frequent itemsets as candidate product features. And combined with compactness rules, redundancy rules and other pruning method to filter candidate product features, the method of mining product features from Chinese customer reviews is proposed. For the non-frequent features the corresponding measure is taken. This thesis further correct the words sequence of phase in candidate features in order to improve the performance of above mining method in Chinese reviews.The Pointwise Mutual Information– Information Retrieval technology is used to measure the degree of semantic association between the candidate features and the product, and to rank the features by the degree to filter out the candidate features which have the relatively low level of semantic association to improve the precision of mining method. Then by analyzing the characters of Chinese customer reviews on travel destinations, some particular change has been made in above mining approach for this special product. The integrated performance of recall and precision has been improved.For integrated the sentiment analysis with product features, this thesis also focuses on modifying and improving on the RWPs (Reference Words Pairs) identification of semantic approach for sentiment classification written in Chinese, and develops the technology of sentiment analysis oriented product features.At last, the thesis realizes the summery information of customer reviews include product features and customer opinion on them. According to frequency of features in reviews and opinion orientation information, features are ranked in the summery. This will make the extracted information more significant which customers are concernedSome products reviews were downloaded from internet for experiments to test these algorithms proposed in this thesis, and some significance difference tests are used to compare results in Chinese and English method. The experiment results prove the validity of the methods.This thesis proposed some new algorithms to solve the critical problems on mining product features from Chinese customer reviews on the internet. These technologies will be expected to help industry and customers more conveniently to access customers'feedback on the product or service. The innovative results of this research will make the application of review mining technologies more widely in the field of e-commerce.