Research On Emotional Classification Of Microblogging Film Criticism

With the development of web2.0, the emergence of micro-blog constantlychanges the way people live. Due to its strong influence and appeal, now more andmore people are fond of publishing movie reviews by micro-blog. Micro-blog moviereview is a kind of audiences’ emotional expression whether the film is good or not,studying this information by sentiment classification not only can help audiencesmake decisions to choose a good movie, but also enable film makers to timely retrievepublic reaction about movies, and then adjust appropriate marketing strategies toimprove the movie box office.Micro-blog movie review is a new model of movie review in the social networkplatform. Now most researches about movie reviews are based on traditional moviereviews. Traditional movie reviews just contains one topic. Micro-blog movie reviewsare different from them.So according to their characteristics, we study the sentimentclassification of micro-blog movie reviews, and our researches are as follows:Firstly, by making statistic analysis on a mass of micro-blog movie reviews, webuilt a movie sentiment lexicon on the base of HowNet, and apply it to sentimentclassification of micro-blog movie reviews.Secondly, we propose a sentiment classification method on the base of topicsentiment sentences. This approach contains three steps: the first step is to extracttopic-related sentences; the next step is to make subjectivity classification, get thesubjectivity topic-related sentences; the last step is to make sentiment classification.Meanwhile, we eliminate zero anaphora for sentences which have the problem of zeroanaphora.Thirdly, we give a semi-supervised learning method on basis of active learning and co-training. Unlabeled corpus of micro-blog movie review can be got easily fromthe internet, but if we want to achieve a large number of labeled corpuses, it needsexpend enormous time and manual labour. With the purpose of decreasing theworkload of manual-label data, in this paper we utilize semi-supervised method, andbring in the idea of active learning within the framework of co-training, thus enhancethe performance of classifier and boost the accuracy of classification.