Along with the proliferation of new media, the user generated content becomes irreplaceable and providing main channel of daily information for people. By get rid of the shackle of the poor information, information technology has entered a big data era. Faced with the data overload, words polarity analysis research appeals the attention of numerous scholars and becomes the important role in national security and information filtering for Internet users, enterprises, and governments. However, due to the rapid change of internet words, the lexicon based sentiment analysis method shows its drawback. Because the traditional method cannot get the polarity of internet words to make the ideal corpus, they usually generate the bad results. This paper presented a topic-based word polarity analysis method which utilizes the LDA topic model and random walk method to get the polarity of a new word. Experiments show that our method achieves the proper accuracy and reasonable results.
