Research On Online News Sub-topics Discovery Method And Its Application In Financial Public Opinion

The rapid development of the Internet makes it more convenient for people to get news and information,so the online news data has grown dramatically.It contains a large number of rich and effective information related to public opinion events,from which sub-topics describing the contents of each aspect of the event can be mined to understand the overall picture of public opinion events from multiple perspectives,obtain the focus of public attention,grasp the trend of events,and provide a basis for the evolution analysis and management decisions.However,the traditional topic discovery methods cannot meet the fine-grained analysis requirements of the current online news sub-topic discovery,such as the inability to find the accurate and effective sub-topic categories from the mass of information,the low degree of differentiation between sub-topics,and the unclear semantics of sub-topic expression.Therefore,it is of great significance to carry out research on online news sub-topics discovery.On the basis of reading the existing domestic and foreign relevant literatures,this thesis analyzes the advantages and disadvantages of various commonly used technologies in the field of topics discovery,and combines the characteristics of online news data to carry out research on online news sub-topics discovery methods.The main works are as follows:(1)Firstly,the LDA topic model is analyzed and in view of its tendency to incline to high frequency words and ignore the low frequency feature words which are representative of the topic,a feature weighted LDA model,LDA-FW,is proposed.And then the model is compared with other traditional methods on Sogou news corpus to prove its effectiveness of this model.(2)Aiming at the problems such as the low degree of differentiation among sub-topics of the same event online news and semantic incoherence of sub-topic keywords,an online news sub-topics discovery method based on LDA-FW model and topic keywords optimization is proposed.On the basis of the LDA-FW model,this method further proposes several optimization steps,including the sub-topics filtering and integration mechanism by filtering garbage topics and merging similar topics,expanding sub-topics keywords by using word vector method,filtering sub-topic keywords based on the idea of network transmission,matching documents and their corresponding sub-topics.The accuracy and readability of sub-topics discovery can be improved by this method.(3)Online news sub-topics discovery research is conducted on financial public opinion events,and fine-grained analysis is made on financial public opinion events.The online news sub-topics discovery method based on LDA-FW model and topic keywords optimization is applied to three types of financial public opinion event corpus,and the advantages of various algorithms are compared and analyzed from the extraction effect of sub-topics keywords and the document sub-topics matching,and the results of financial public opinion sub-topics discovery are analyzed.The experimental results show that,compared with other sub-topics discovery methods,the online news sub-topics discovery method presented in this thesis has a better online news sub-topic discovery effect under various indicators of relevant evaluation.This method can effectively improve the quality of sub-topics discovery and provide reference for the research of sub-topics discovery...