Research Of Sentiment Analysis For Chinese Product Reviews

Rapid advances of Internet technology making e-commerce websites develop fast,more and more people get the product reviews through the network platform and guide consumption based on these information.Data mining and corresponding analysis to these exponential growth of information for identifying the emotional tendency can not only help us understand the users' satisfaction and views on commodities,but also provide a reference for the business decisions,thus improving the quality of goods and service attitude.This paper firstly introduces the flow of Chinese text sentiment classification and related core technologies,then correlated analysis and research are made from the two aspects of product reviews obtainment and product reviews sentiment analysis methods respectively,and the paper uses a machine-learning method and a dictionary-based approach to product reviews' sentiment analysis.As to emotional analysis of comments,firstly,corpus data need to be obtained from the corresponding shopping platform,reviews of different categories of goods are usually concentrated in one sector,and the pages are highly structured.Based on this,the paper uses the web crawler to process web pages in real time and obtain comment information,thus to get the comments data for emotion classification.Then,comments emotional analysis is made according to two different ideas.In dictionary-based method,an integrated emotional dictionary for product reviews information is constructed,where the basic emotion dictionary uses “sentiment analysis set with the words” released by HowNet,while the network emotion dictionary is mainly based on existing incomplete network dictionary,by collecting and annotating the emotion words from comments,the network emotion dictionary is expanded.Furthermore,SO-PMI algorithm is employed to calculate the tendentiousness of emotional words for the expansion of emotional dictionary.Finally,according to the phrase structure,emotional tendencies weighted sum of feature words in comments is made to obtain the entire comment tendencies and emotional strength.In machine-learning method,a Bayesian classification model is created,and a hybrid feature extraction method is presented,which is conducive to retention of classification characteristics,and deletes the redundant features,thus improving sentiment classification results.At last,four categories of reviews data are acquired from TMALL as test corpus,that are,digital products,clothing,foods and books,then contrast experiment are conducted by using the two methods in the paper.The results s how that the two methods are reasonable and the machine-learning method has better classification quality.