Sentiment Classification By Combining Lexicon-based And Machine Learning Methods

In recent years, with the development of e-commerce, SNS and micro-blog, the internet entered a new era. With the production of the user generated content, which marks that, the people is no longer simply an audience, but has become part of the internet. For what, they have the space to express their views. There are so many views now. These vast amounts of unstructured information is clearly contains a great deal of information. Companies need to obtain the views of users of the products. The government needs to know the people reflect on a policy. How to deal with the information to gain the knowledge what we want is the current focus of attention of scholars. Opinion mining and sentiment classification is a new area focus on deal with this problem. It separates the views to two parts, which are positive and negative, according to the emotion of the writer. With the help, we will know that the emotion of the audience expressed by the text for or against. And a product is recommended or worthless.In this dissertation,the problems of text sentiment classification on document level are investigated.The main contributions of this dissertation are summarized as follows:Firstly, we propose a new self-supervised model for sentiment classification. In this model, we combined lexicon-based method with corpus-based method to address the major drawbacks of only using one of these two methods. The former does not adept well to different domains, while the latter one requires much effort of human annotation of documents. Our self-supervised model can overcome these drawbacks. Secondly, we improve the TFIDF model and used it into the SVM classifier. The result proved that this method is move efficient.