Research And Application Of Financial Text Emotional Analysis Technology

The financial market is a highly complex system,and investors need to constantl y pay attention to relevant information in order to analyze and predict market trends,r educe risks,and increase profits.With the development of internet finance,financial i nformation is more frequently presented in text form through the internet,and these te xts contain the authors’ sentiment towards the development trend of the financial mark et.According to the theory of behavioral finance,emotional information in online text s can affect investors’ psychology and spread through behavioral patterns such as "her d behavior",which can then affect the development of the stock market and even lead to the outbreak of financial crises.Therefore,analyzing the sentiment in financial text s and controlling market sentiment is of great significance for both countries and indiv iduals.Among existing sentiment analysis methods,dictionary-based analysis methods are costly and have poor applicability,while machine learning-based methods require manual feature extraction with professional knowledge and experience,and are difficu lt to achieve effective results on financial texts with ambiguous sentiment,large data v olume,and fast updating speed.Therefore,this paper proposes a financial text sentim ent analysis model based on deep learning technology,and has carried out a series of work around this model.This paper mainly studies the following:(1)Design of a topic crawler based on BERT.To solve the problem of a lack of h igh-quality sentiment analysis datasets in the financial field,this paper proposes a topi c crawler based on BERT,which uses a deep learning model to judge the relevance of webpage content to the topic,filters out webpages that are relevant to the topic,and b uilds a financial text sentiment analysis dataset containing 15,000 data points based o n the filtered webpages.(2)Construction of a financial text sentiment analysis model.To address the prob lem of insufficient feature extraction of a single deep learning model,this paper propo ses a feature extraction model based on MHA-Bi LSTM-CNN,which combines the lo cal feature acquisition ability of CNN,the sequence feature extraction ability of Bi LS TM,and the key feature extraction ability of multi-head attention mechanism(MHA).The model achieved an accuracy of 85.9% on the constructed dataset and outperform ed other models.(3)Development of a market sentiment monitoring system.Based on the first tw o research contents,this paper developed a market sentiment monitoring system that p rovides users with personalized market sentiment visualization analysis to help users quickly understand the market.The system’s backend uses the Flask framework,and t he frontend is divided into two subsystems according to requirements: the market sent iment display subsystem for ordinary users,which is integrated into client trading soft ware and developed using QT;and the sentiment labeling subsystem for administrator s,which is accessed through a browser and developed using Vue.