Design And Implementation Of Short News Sentiment Analysis System For Financial Field

The sentiment information of news has important reference value for the financial field.Short news is more refined,especially the news flash can reflect the key information.With the change of short news communication carrier from paper media to Internet,the financial institutions began to try to provide the sentiment information of short news to specific business scenarios.At present,there is a lack of sentiment analysis data set of financial news,therefor,there is a lack of corresponding sentiment automatic tagging method,and the existing products that provide news sentiment information can not meet the needs of this system.Therefore,this system aim to help labeling the sentiment information of financial short news and assist the financial short news analysts to analyze and summarize.Firstly,this thesis introduces the theoretical basis of automatic labeling of sentiment information of financial short news.Then,to achieve the goal,the sentiment analysis data set of financial short news with more abundant news types,which contains labeled financial short news with sentiment tendency and the dictionary of financial short news,is constructed.Then a financial news classification tree is constructed through keyword extraction,named entity recognition and reference from various sources.The construction also supplements the dictionary of financial short news.On this basis,according to the needs of financial short news analysts and financial short news labelers,the demand analysis of the system are carried out.Then,the system is designed and implemented in detail,including data update,information display and manual labeling.To update the data automatically,this system realized the news acquisition,sentiment pre labeling and category pre labeling.For sentiment pre labeling,the automatic labeling method of financial Short News sentiment tendency is explored,which include the method based on Naive Bayes algorithm and the method based on BERT,and finally chooses the method based on BERT to apply to financial Short News sentiment tendency pre labeling.In the manual labeling module,realized news display and news labeling for financial short news labelers.In the detailed design and implementation of information display module,realized sentiment tendency browsing,filtering and sorting,data statistics display for financial short news analysts.Finally,the system has passed the test of the test case.