A Research On Feature Selection And Opinion Classification Of Chinese Web Product Comments

With the development of Internet technology, the dissemination and sharing of information become more and more convenient. Nowadays, People are able to present their evaluation of products either in commercial websites or they can express their opinions or views about almost anything in the forums, blogs, social networks and other media. While it makes information sharing more convenient and efficient, it also provides a large number of users'opinion information on the web which leads to more difficult for people to obtain useful information. Therefore, to use of existing information technologies to mining these comments and get access to valuable information is of great significance. This paper discusses these problems and mainly completes the following work: to research the existing product opinion mining theories and technologies of web comments thoroughly and delicately, from research categorization to technical support, from research framework to related algorithms, and does comparison and analysis; Based on present research, does two experiments separately, first use the traditional text classification method only to do experiment and then add an opinion vocabulary to traditional text classification method to do experiment, with machine learning method, paper assesses these two methods; From the semantic point of view, to combine the frequent pattern extraction and information retrieval PMI algorithms to extract product features and does empirical research, conclusions and suggestions are given; Build an analysis and design framework of the integrated product feature extraction and opinion classification system, give the mining results to the user in the form of visual display, to help users do decision-making. This paper follows strictly with the design science methods, the proposed method in this paper are verified by experiments, and the results are analyzed. Compared to other Chinese product reviews mining research, this paper has three innovations: (1) Present the approach of building opinion vocabulary to do classification, to optimize the results of application of traditional text classification methods in opinion text categorization. Given the hotel reviews as a sample, with the machine learning method, by constructing a small field of opinion lexicon, to optimize traditional text classification; (2) Starting from opinion mining level, to propose a feature-level product reviews classification, mining frequent features by association analysis method, with the method of semantic PMI to correct the mining results; (3) Combining feature extraction and opinion classification , to put forward the integrated system analysis and design framework, does opinion categorization according to extracted features, improving the classification accuracy, and to show the visualize pictures to users.