Research Of Opinion Review Mining Based On Semantic Comprehension

With the rapid development of Internet technology, the number of texts on the web is increasing rapidly, therefore how to mine useful information from these vast texts becomes an important research subject. Based on the research of text mining and combined with the popular e-commerce and reviews on web, this paper mines the information from product opinion reviews, and then feeds back to the merchants and customers.In the first part, this paper detailedly introduces the development and relevant technology of text mining and opinion review mining, and then systemically analyzes the technology and algorithm of text mining in the second part. At the same time, it also deeply researches on the important branch—text categorization, and discusses the general process of text categorization, In the third part, this paper designs two kinds of improved text categorization methods for text preprocessing. Because traditional rule matching is too complicated, to optimizing the complexity of algorithm, the first method is attribute reduction based on rough set; and the second one is text categorization method based on term co-occurrence concept, which can make up the independence of traditional terms and express the semantic information of text by term co-occurrence concept. The experiment results indicate that the two models all improve the result of classification.In the last three parts, this paper makes some research on mining product review based on the popularity of internet review and text mining theory. Under the help of some research results of foreign scholars, it also designs two models of mining product review. The former one is mining product features based on association rule and semantic comprehension, and the main work of it is how to mining product features.The whole process is relatively simple, especially depending on statistic and lacking of detailed semantic analysis. According to the shortage of the former model, this paper makes some improvment at the last part. The last model uses dependence relation to find opinion words, importes semantic similarity to product review, integrates the difference between Chinese and English, and enchances both the recall and precision. In sum, because of the freedom and particularity of internet review, the precision is better than recall.