Analysis Of Stock Forum Mining Baseds On Deep Learning Method

The stock market usually receives the favor of the majority of investors with its high profitability,but the unpredictable changes also make investors unpredictable,so it inevitably causes the uneasy mood of investors.In the era of rapid development of Internet finance,the stock forum emerged as The Times required,and is gradually becoming an important platform for investors to exchange experience in stock trading.With the development of deep learning methods,sentiment analysis,text classification and other methods using natural language processing tools have become a popular trend of forum data analysis.Through literature study,this article understands,internal active BBS with numerous real investors,their mood changes reflect the BBS of each stock investors operate wind direction,if you can learn that investor sentiment information in time,can be for the government supervision department,market intermediaries,and provide decision-making reference for listed companies,investors and other subject,Promote the healthy development of China’s stock market.In order to deeply explore the internal changes of the stock market and accurately measure the community theme of the financial forum,this paper selects the stock forum of Oriental fortune as the research object,and uses python3.7 to crawl the Posting data of the leading stocks of seven industry sectors in the stock forum of Oriental fortune.Four stock forums,vanke A Bar(000002)in the real estate sector,Sany Heavy Industry Bar(600031)in the machinery sector,Shere Wine Bar(600702)in the wine industry and Shaanxi Black Cat Bar(601015)in the coal sector,are selected as the general model collection objects.Hengrui Pharmaceutical(600276)in the medical sector,INDUSTRIAL and Commercial Bank of China(601398)in the banking sector,and Zhonggong Education(002607)in the education sector are taken as the collection objects of the optimization model.The text data are cleaned,and the five categories of investor sentiment is-5 are proposed based on the theoretical research of network forums.The emotions in the text were divided into five categories:aggressive fighting,positive optimism,neutral caution,negative pessimism and anger and despair.A total of 12,893 data were annotated manually according to the requirements of model construction.In order to build a high-quality classification model of investor sentiment in stock forum,this paper applies the natural language processing model of deep learning to compare the performance of BERT,ERNIE,Bi LSTM and Bi LSTM+Attention models with outstanding performance in current text classification tasks based on a large number of labeled data sets.Finally,by comparing evaluation indexes,BERT text classifier,which is more mature and stable,is selected as the general model of investor sentiment classification in stock forum.This article choose by adding three new shares in turn industry groups data,investor sentiment classification model of iterative process,to reduce to join the sample size,the purpose of reduce the training time,form a stronger generalization ability of stock BBS classification system,Oriental wealth network shares BBS of investor sentiment classification model was set up and optimize.Finally,in the stock forum scenario,the ISs-5 five classification radar chart IS used for visual representation,and three distinctive investor sentiment radar charts of Anhui Huri(600761)in the machinery sector,Kweichow Moutai(600519)in the wine sector,and Shanghai Pudong Development Bank(600000)in the banking sector are selected to compare with the stock prices of the month.It is found that there is a large correlation,so as to realize the classification application of investor sentiment in stock forum.