Research On Topic Evolution Based On LDA Model And AP Clustering Algorithm

With the rapid development of the Internet, network information are showing the trend of explosive growth, and because of wide coverage, spread fast and strong affinity characteristics, Internet news become the main way for people to obtain information. Therefore, establishing the news topic evolution analysis system can help users get more valuable information from the massive data of Internet news, analysis the development trend of news events, have very important role on public opinion monitoring for the government and business mining for enterprises.Firstly, this paper deeply discussed and studied the background and significance of news detection and evolution model; secondly this paper studied the topic modeling analysis in the news evolution detection application based on the theory of co-occurrence words theory and LDA topic model; then for the single topic evolution and threshold setting, combined with LDA topic modeling theory and cluster theory, LDA-AP topic evolution model is constructed. At last, the paper introduces the background and current research situation of the adaptive topic evolution model, combined with the adaptive adjustment strategy based on LDA topic model and the time decay of the mining news topic relationship among different time windows. On this basis, completed the topic evolution experiment by collecting the news data from websites such as Sina and NetEase, by comparison with a reference method, the feasibility of the LDA-AP topic evolution model based on the time decaying, proposing the existing deficiencies and follow-up research in this paper.Through the experimental study found that the LDA-AP model based on time-decay topic evolution analysis method could not only mining multiple news events relating to the topic evolution trend over the time, but also improve the accurate rate of topic evolution, effectively improve the performance of the topic evolution analysis.