Research And Realization Of Text Affective Tendency Analyzing System Based On Support Vector Machine

Nowadays, we are in an Internet era of information explosion. With the wide spread of the internet, more and more people's lives get affected by the net, especially with the emerging of Web 2.0 era. we can not only have access to incredibly massive information online, but also post blogs, micro blogs and comments, etc., which lead to the explosive growth of information on the net. This huge amount of information makes internet a huge repository of information, which has brought great convenience to people's lives. The huge amount of information online poses a problem: how to get the needed information from the internet quickly. This becomes an urgent problem in the field of information processing. The emergence of search engines, to a certain extent, meet people's needs of getting information quickly, but with the growing of information on the Internet, people have higher and more specialized requirements on the accuracy of the search. Text affective tendency analyzing has become one of the current requirements in information searching.There exits certain emotional tendencies in the news, blogs and comments on the Internet. If we extract these emotional tendencies and classify the related articles, it will facilitate the decision-making of individual users, companies and the government. For example, for individual users, when we want to buy a product, they desire to acquire a comprehensive and clear understanding of the product before they make the decisions: whether to buy it online or in the store. Therefore, if we classify the online reviews of this product according to the emotional tendency, they will be offered the chances of understanding the satisfactions and the dissatisfactions of this product. In other words, the users can identify the advantages and the disadvantages of this product clearly and fast. For corporate users, the companies want to know whether it is popular or not for its products or services. We could classify the related reviews based on the emotional tendency, and thus companies will know the advantages and disadvantages of its products or services, whose effect is equivalent to that of a satisfaction survey. For government as all enacted policies should be founded on public opinion, we can classify the emotional tendencies of the online comments when a intended policy is made public, thus helping the government have an access to the public opinions in its decision-making process.In this essay, the Text Affective Tendency Analyzing System is introduced, which is designed by Shenzhen Key Laboratory of Intelligent Media and speech, based on the Business Mining and Service System (BMSS) and the Affective Tendency Classification research at home and abroad. It is also the opinion tendency analyzing module in BMSS. We implemented the sentence affective tendency classification algorithm which was submitted to the 2010 proceeding of NTCIR-8 by WIA team. Compared with the other algorithms proposed in NTCIR-8, this algorithm proves to have better performance. At the same time, we make an improvement to this algorithm, the improved algorithm greatly enhance the time efficiency at the expense of a little lost of accuracy. Based on the improved sentence affective tendency classification algorithm, we designed and implemented a text affective tendency analyzing system, which makes it possible to adjust the threshold of the polar sentence proportion to control the degree of the text affective tendency. At the same time, we propose a method to calculate the degree of the text affective tendency. In performance, the system in this essay meets the need of BMSS well in the time efficiency and accuracy rate. The essay also raises some problems which need further research and improvement.