Construction and application of a Web text-oriented integrated sentiment feature library mined by a big corpus
Meijuan Liu 1 , Shicai Yang 2
COMPUTER MODELLING & NEW TECHNOLOGIES 2014 18(12A) 370-376
1 School of Foreign Languages, Zhejiang Ocean University, Zhoushan, 316022, China
2 Ningbo University of Technology, Ningbo, 315016, China
This paper describes the implementation of a system containing a web text-oriented integrated sentiment feature library (hereinafter referred to as WTISFL) and its application in sentiment analysis. Sentiment library plays an important part in sentiment analysis. A quicker and more complete method of constructing new sentiment library is presented in the paper. Firstly, the structure of WTISFL which sentiment analysis needed is proposed. Besides, WTISFL is mined from the big corpus database. Moreover, our sentiment word set is extended on the basis of existing sentiment resources, semantic similarity calculation of HowNet and computation of Chinese Synonym Thesaurus. Finally, the WTISFL is checked manually. Based on the above WTISFL, Web texts are studied from the perspective of sentiment analysis with the method of maximum entropy classifier. The experiment shows that WTISFL in this paper is extremely effective in sentiment analysis, which can evidently improve the performance of web texts sentiment classification.