Research And Application Of Financial Public Opinion Analysis Technology

With the rapid development of Internet technology and financial industry,network public opinion has gradually become an important factor affecting the stable development of Chinese Enterprises.The Internet public opinion has a strong social influence and spreads very fast.The public opinion analysis of financial public opinion reports is conducive to the relevant institutions to easily understand the public opinion of events and make correct guidance and control,as well as the sustainable development of the financial market.This thesis uses the topic model,emotional dictionary construction,public opinion analysis and other technologies to conduct detailed research on financial-related network public opinion and design a financial public opinion analysis system.The system can effectively identify the emotional words in the text from the improved topic model,matchs the emotional words with a more comprehensive and perfect emotional dictionary in the financial field,and calculate the emotional tendency value of the emotional words,so as to more accurately classify the financial text public opinion.The content of this article is as follows:1.Improved WHDP model.The traditional HDP topic model is based on the "bag of words" hypothesis,which causes confusion in text semantics.To address this problem,this thesis proposes a window-based hierarchical Dirichlet process(WHDP)topic model,that is,introducing a window mechanism into the HDP model,using the window to divide the document into smaller fragments,and moving the window to guarantee the order relationship between words and reduce the semantic confusion of text.Experiments show that whdp model has the advantages of less perplexity and stronger generalization ability.2.Improved CHDP model.The HDP topic model treats the document as a simple combination of word frequency vectors,resulting in a lack of text semantic information.To solve this problem,this thesis proposes a Centroid-word based hierarchical Dirichlet process(CHDP)topic model.The model is centered on the word to be calculated,expands several words before and after as a window,and then the topic probability of each word is calculated for each window.This method ensures the orderness between the windows,thereby ensuring local ordering between words.The experimental results show that the average perplexity of the CHDP model is less than the WHDP model.3.Construction of financial public opinion analysis model.Aiming at the problem that general emotion dictionary can not be applied to public opinion analysis in specific fields,this thesis selects the top 100 words with the highest frequency as the seed words,and then uses SO-PMI algorithm to calculate the emotional similarity between the candidate words and the seed words,and constructs a financial sentiment dictionary,and then merged the financial sentiment dictionary and the basic sentiment dictionary into a more comprehensive financial sentiment dictionary.Finally,the WHDP / CHDP model and the sentiment dictionary in the financial field are used to classify public opinion in financial texts.Experiments show that this method has a higher accuracy of public opinion classification.Finally,based on the above research content,this thesis implements a public opinion analysis system based on the financial field,and achieves the purpose of practical application.