Research Of Sentense-Level Sentiment Analysis Based On Association Rule And Graph Ranking

Nowdays, Internet is playing an irreplaceable role in our work and life. It has become a part of our lives. The emergence of new websites make the information of internet increase fastly, through blog or microblog and other these new internet platforms, people express their opinions or comments on hot news or social things, this makes the proportion of subjective texts increase dramaticly. What methords can be used to mine the authors'views from the subjective texts? This has become a hot issue in the field of natural language processing. It is very difficult to resolve so much information simply by relying on artifical methods, but the sentiment analysis precisely solve this trouble. By computers'analysis and processing, we can know the authors'views, that is called sentiment analysis.We research the sentence level sentiment analysis in this paper. We first analyse the development of sentence level sentiment analysis in the world and the bottleneck which it encounted, and then we promote the idea of sentiment analysis based on association rule to solve the problem of ambiguous words and propose the method of graph ranking to deal with the problem of muti-affect words, and that can provide help for future study.First, we propose the method sentiment analysis of sentence based on ambiguous words in the enviroment of association rule's classic application named "shopping basket problem" We use association rule to dig ambiguous collocations in the sentences of corpus COAE2008, apply the method of mutual information probability to filter the noises, then we determine whether the collocations have sentiment, after that we will built a ambiguous collocations dictionary on which we determine sentences' tendency by the method of senmantic analysis. Experiments we worked on the COAE2008 corpus showed that ambiguous collocations is very important and essential on the sentiment analysis, it can make a large influence on the accuracy of sentiment analysis.Second, according to the successful application of pagerank algorithm in the aspect of link analysis, we propose the method called word emotion disambiguation based on graph ranking. We preprocessing the corpus at firsrt, and then built the graphs about the meanings of words through modern chinese dictionary. Aftering builting graphs.we use the algorithm to do random walking and iterative computing on the graphs, and select the max iterative value amoung the muti-affect word's meanings as right result. Last, we also do some comparative experiments on the corpuses of microblog and lexicon ontology, the results can demonstrate the feasibility of the method full).At last, we combine the ideas of association rule and graph ranking, and analyse the sentences of COAE2008, the evaluations of the experiments all be improved, and this reflects the significance of our research.