The Research Of Public Opinion Analysis Technologies Based On Machine Learning Theory

With the development of network technology and application, our country has nearly 650 million Internet users in December 2014, the Internet is becoming one of the important ways to spread information. Relative to the traditional way of transmission, the Internet is an important way to lower the threshold of the information dissemination, giving the majority of Internet users a convenient way of communication between the users.With the development of China’s securities market, more and more investors to join. The Shanghai and Shenzhen a-share accounts up to 185 million at March 2015. At the same time, along with the application of network technology in securities trading, investors’trading strategies can be conducted by the information from the Internet. It brought convenience and challenge to investors at the same time. On the one hand. a lot of information on the network can help investors make trading decisions. On the other hand, the complexity of information exists on the network is hard to make rational and objective investment decisions. Therefore, the research on network public opinion is not only the supplement of behavioral finance, but also the improvement of the market efficiency theory.According to the real application, using the Java language and open source tools web crawler technology and the lucene index to develop information retrieval system of professional field at first. On this basis, we research topic detection technology, emotional tendency analysis and forecast the China’s securities investor confidence index. Because of traditional topic model (LDA) cannot accurately capture the article structure and distinguish the similar topic in the same field, an improved model is proposed. Because of the traditional sentiment classification algorithm can’t understand compound semantic, this paper adopts a kind of advanced deep learning algorithm to analysis the Shanghai and Shenzhen a-share market reviews. Because of the traditional support vector machine (SVM) regression algorithm for noise produced over-fitting, we improved a new support vector machine regression model(ASVDD) to forecast the China’s securities investor confidence index to guide investment behavior. Through experiments, the algorithms can effectively improve the effect of public opinion analysis.