Research On Sentiment Analysis With Internet Product Review

With the rapid development of Internet and e-commerce, more and more people get used to shopping online. At the same time, people are troubled by a large number of Internet products and product reviews when choose cost-effective products. As a result, the sentiment analysis of Internet product review is particularly important. The sentiment analysis is mainly for the field of Internet products. In order to found people's views and opinions of special Internet product, using the method of machine learning analysis its emotional tendencies automatically.The research topic is research on sentiment analysis with Internet product review. The main purpose is to help consumers found appropriate products and help producers have a better understanding of the products. The method is to use computer technology to analyze large-scale Internet product reviews and get people's attitudes. The main work of this paper includes the following aspects.Firstly, complete the pretreatment work. Select a special electronic product as the research object and get the product reviews. We should process the data before perform an analysis, the work mainly includes Chinese word segmentation, filtering, tagging, data cleansing and data classification.Secondly, study and improve the feature selection process. Feature selection play a decisive role on emotional classification. Appropriate features can improve the accuracy of classification. On the one hand, for different text features including emotional words, adjectives, adverbs, verbs, and emotional mood words. I raise that emotional mood words have a good assistant effect on emotion classification. On the other hand, compare the feature selection methods, point out the deficiency of traditional mutual information method and put forward an improved mutual information feature selection method based on the ratio of positive and negative. Through experiments we find the optimal combination of speech features and the improved mutual information feature selection method with better performance.Finally, analyze the Three-way Decision and introduce into the emotional classification. Three-way Decision has better performance when dealing with uncertain things. This paper proposes a multiple decision weighted hybrid classifier based on Three-way Decision and gives the main idea, relevant rules and definitions. Using NB classifier and SVM classifier to make three-way decisions respectively with optimal threshold, the classification of the last texts in border areas is determined by NB classifier and SVM classifier. Experiments prove that multiple decision weighted hybrid classifier has certain advantages and can help to improve the accuracy of classification.