Research On Comparative Opinion Mining Based On Web

Accompanied with the trend of "civilian's Internet", the users whom received information from the web passively in the past are becoming the information's active creators gradually, so there are a large number of valuable user-generated-comments on the web. How to mining the concerned points of the potential consumers from the massive data effectively and conveniently has been being an urgent task. Thus, the opinion mining technique emerged and became a hot spot in the information processing field rapidly.In order to obtain detailed results of opinion mining, this paper aims at opinion mining problem on the sentence level, the major work of this thesis include:(1) Summarize the related tasks(task3 and task4) in the second Chinese Opinion Analysis Evaluation(COAE2009) and our method proposed about them, analysis the results of evaluation, after that, the core elements of an opinion is proposed, and the main goal and task of opinion mining at the sentence level is defined.(2) Research the problem of semantic role labeling for Chinese comparative sentence. By referring to the syntactic structures of comparative sentences, the paper constructs the original hybrid comparative patterns, then generalizes them to improve the accuracy and coverage rate of the labeling results. Further, a two-level algorithm is designed for comparative entities labeling and comparative feature labeling respectively. The experimental results indicate effectiveness of the proposed method.(3) Mining opinions from the Chinese comparative sentences. A description form of comparative opinion is firstly proposed, which normalizes the different types of comparative results and provides a basis for making induction and integration between these results. Then, construct comparative patterns according to the syntactic structure of Chinese comparative sentence and groups the patterns by the sentiment information carried by them. Finally, on the basis of the results of semantic role labeling, an algorithm based on comparative patterns for mining opinions in Chinese comparative sentences is proposed. Finally, the performance of the algorithm is tested on real corpus.(4) Construct a comparative opinion mining system based on web automobile comments. The system integrates the determining,Semantic Role Labeling and opinion mining of Chinese comparative sentences. Update the database by using the web spider regularly, after the preprocessing of part of speech tagging, label the identified comparative sentences and mining comparative opinion, finally show the detailed contrast between two different automobile brands questioned by the users.