Text Emotion Analysis Technology Based On Semi - Supervised Machine Learning

With the rapid development of Internet technology, people who accept website information passively in the past become to distribute, transmit and share information on the Internet actively. Therefore, followed by a lot of user information and the diversification of the network information. Facing with quantity of information and data on the Internet, on the one hand, users need to spend a lot of manpower and material resources to judge and identify, to gain their valuable information, to help their daily life. On the other hand, producers and manufacturers hope that through these vast amounts of data to analyze the mass of a product or service evaluation and feedback, to help them make the right decisions in a rapidly changing situation and judging. Sentiment analysis technology arises at the historic moment, it can effectively retrieve information network comments, this paper studies mainly aimed at the evaluation objects tendency of emotional problems.First of all, in this paper, the sentiment analysis technology research status at home and abroad, introduce the technology used in the field of sentiment analysis and its steps and main support algorithm, and the various algorithms are analyzed and compared. The algorithm of sentiment analysis need to have a certain number of labeled samples that used for training, and the accuracy of sentiment analysis will increase with the increasing number of labeled samples. Through the network we can get a lot of unlabeled text, if you try to label and choice these unlabeled text, it will cost a lot of material resources and manpower. So we introduced a semi-supervised machine learning methods, try to use the unlabeled text, the implicit information of unlabeled text is added into the label samples to improve the classification performance of the classifier. Because of some disadvantages of semi-supervised machine learning methods, in the case of fewer labeled samples, the performance of the classifier is often not enough ideal, lead to the higher error of picking the sample, at last can not significantly improve the accuracy of classifier. Then, the paper detailed the current academic circles at home and abroad on semi-supervised machine learning problem of improving methods and opinions, based on summary and induction of them, we raise the algorithm in this paper.Finally, the paper puts forward a kind of semi-supervised sentiment classification method based on dual system. The method on the basis of the existing dual system, transplanting it to semi-supervised learning, in the process of choosing samples, introducing method and redefine the rule of confidence and differences to improve the quality and accuracy of samples, so as to improve classifier learning rate, eventually get high accuracy. The performances in the last show that the proposed algorithm on existing semi-supervised learning algorithm obtain a certain increase and have the validity and feasibility.