Application Ot Topic Model In Evolution Of Financial Texts

The research in the economic field is inseparable from the support of data,while the extraction,analysis and use of related unstructured data is a subject to be further explored.Combining the data mining method with the unstructured data information in the economic field is of great significance to the development of data mining methods and research in the economic field.Based on the stock news,this article focuses on the retrieval of topic information in the financial text,using more than 170,000 news related to SSE 50 Index constituent stocks as corpus,exploring the problem and the improvement of the application of topic model in financial texts.This article uses the dynamic modeling method to analyze the evolutionary pattern of the topic in the financial text.The research shows that the topic model can extract the topic information effectively in financial text.The main contribution of this article is to:(1)Through the statistical analysis and topic modeling in large-scale corpus,this article sums up the characteristic of topics in the Chinese financial text:there are two types of special topics in the topics of individual stock financial news:core topics and event topics.The core topics throughout,can be used to observe changes in the main business of individual stocks;event topics are related to events,can be used to help identify the events related to individual stocks and their changes.;(2)In the setting of topic number,this article proposes a method for the determination of topic number based on the minimum redundant topic.Redundancy topic is useless.This method searching the max topic number without redundancy topic by dichotomy,is easy to understand and fast;In the topic similarity measurement problem,by analyzing the differences between existing similarity measure methods and the habit of human while reading the topics,this article proposes a measure method based on the weighted rank correlation coefficient,which has a good capability in topic semantic similarity measurement.