Research And Application Of Topic Evolution Model Based On LDA

With the wide application of Internet technology,the explosive growth of data resources brings serious information overload problems.Mining topic evolution information in text data can help people get information from massive amounts of data more efficiently.The LDA-base topic evolution model analyzes the text data by combining time characteristics.As a result,it can find the hot topic in time,reflects the evolution trend of the hot topic and provides a good way to solve the information overload problem.This paper studies the model of fine-grained topic discovery and evolution,aiming to analyze the evolution of targeted topics from time series text.Different from the traditional coarse-grained topic evolution model,this paper studys the evolution trend of the topics of interest to people based on an extended model TTM of the LDA topic model.The main research results of this paper are as follows:1.This paper proposes a TTM-based pre-discretized topic evolution method.This method can analyze the evolution of the target topic content and use effective topic association filtering rules to solve the problem that the topics are difficult to align on different time windows.2.This paper proposes a TTM-based post-discrete topic evolution method.This method can model the targeted target topic and reflect the density evolution trend of the target topic.3.This paper proposes a embedded target topic evolution model TTOT.This model uses the Spike and Slab prior to sparse modeling of the target topic,and uses the time variable obeying the beta distribution to analyze the change of the target topic with time.Therefore,the TTOT model can perform target topic discovery and its evolution analysis simultaneously without the need for additional discretization of text data by time characteristics.The methods proposed in this paper can filter the irrelevant content and analyzethe dynamic evolution of the target topic through keywords.The numerical experiments based on NIPS dataset and CNKI dataset show that the three topic evolution methods proposed in this paper can effectively identify the target topic and give the evolution map of the content or intensity of the target topic.