Research On Sentiment Classification Based On Co-training In Semi-supervised Learning

With the rapid development of Web2.0, the Internet has produced a large number of user generated contents. These user generated contents contain a lot of useful emotion information and have an important value for the individual’s decision-making, enterprise product improvement, and so on. Therefore, how to use the sentiment classification technology to detect the emotion information in the mass user-generated contents has become a hot area in academic and industry. At present, although machine learning based sentiment classification methods have achieved good results, it require a lot of human efforts to obtain labeled examples in real world application. On the contrary, it is easy to get unlabeled examples. Therefore, how to use a large number of unlabeled examples and a small amount of labeled examples to learn has become one of urgent research problems in the area of sentiment classification.This research proposed a new sentiment classification method based on co-training in semi-supervised learning to solve the above problems. Firstly, this research analyzed the research status of sentiment classification and semi-supervised learning, and clarified the research problems and future directions. Secondly, this research systematically analyzed the basic theory of sentiment classification and semi-supervised learning, including major tasks of sentiment classification, major methods of sentiment classification, basic hypothesis of semi-supervised learning, validity of semi-supervised learning, major methods of semi-supervised learning, and so on. Then, based on the above analysis, this research studied the construction approach of sentiment classification model based on co-training in semi-supervised learning. For balanced dataset, this research constructed IDSSL based sentiment classification model. For imbalanced dataset, this research constructed hybrid strategy based sentiment classification model. Finally, this research applied sentiment classification method based on co-training in semi-supervised learning into practice. Two application scenarios, i.e., electronic commerce and medical social media, were selected to prove the effectiveness of two sentiment classification methods based on co-training in semi-supervised learning. Experimental results indicated that the proposed methods in this research achieved the better results under the condition of different data distribution.Through this research, on the one hand, the semi-supervised learning method was introduced into the sentiment classification problem. The basic theory of sentiment classification and semi-supervised learning has been expanded, and on this basis, the sentiment classification models based on co-training in semi-supervised learning have been built. On the other hand, this research applied sentiment classification model based on co-training in semi-supervised learning into specific practical problems, which expanded the application scenario of sentiment classification and semi-supervised learning.