Study On Sentimental Polarity Shifting With Its Application

Along with the rapid development of Internet, the network has become the main wayfor consumers to feedback their views. The shared comments on the network not onlyprovide an information display platform for businesses, but also let consumers exchangetheir experience. Under this circumstance, it needs an effective means to organize a varietyof large amounts data, and show the statistical results in a intuitive way. In this context, thetask of sentiment analysis emerges.Sentiment classification is one of the importment research topic in the study ofsentiment analysis, which aims to classify a text according to some sentimental categories,such as positive and negative. Nowadays, sentiment classification has become a hotresearch issue in Natural Lanuage Processing. However, the sentimental expression of textis complex and the judgment of the polarity is often not ideal. Among them, a major causeaffects sentiment polarity classification is the phenomenon named sentimental polarityshifting, which means the polarity of a sentimental word has changed or reversed becauseof the impact of other word or phrase. This paper conducts extensive studys on sentimentalpolarity shifting, with the efforts and goals on:First, this paper analyzes the phenomenon of polarity shifting and build relavantcorpus. Specifically, by observing corpus, we focus on some kinds of linguisticphenomenon of polarity shifting about the text in the corpus, and propose the architectureof most polarity shifting structures including: negation, contrast transition, modality, andimplication. In this architecture, we make a detailed annotation of polarity shifting incorpus, and marke the different types of trigger words of polarity shifting aboutsentimental words. Given the annotated corpus, we give various types of statistical analysisand comparative study.Then, this paper presents automatic detection methods for detecting polarity shifting.According to the structure of polarity shifting and trigger words summed up, we proposetwo ways of automatic detection: rule-based detection methods and machinelearning-based detection, where rule-based method apply of the appropriate trigger words table automatical detected the different structures of polarity shifting, machinelearning-based detection empioys the annotated corpus of polarity shifting to train amachine learning-based classifier. Experimental study demonstrates the effectiveness ofthe proposed detection methods.Finally, this paper applys the detection of polarity shifting into the task of sentimentclassification. Given a sentimental dictionary and the results of polarity shifting detection,we conduct sentiment classification by considering the polarity shifting in twoclassification methods: termcouting and bigraph-based classification. The experimentalresults show that considering the situation of polarity shifting could significantly improvethe overall performance of sentiment classification.