Research On Stock Forecasting Methods Integrating News Topics And Comment Sentiment Feature Analysis

The stock market affects all aspects of social life.As an important part of the economic market,it is now receiving widespread attention from domestic and foreign investors.Therefore,studying the changing trend of the stock market is of great significance to the national government,corporate institutions and individual investors.The stock market is not static.It has the characteristics of volatility and unpredictability,and is closely related to national policies,economic development level,social environment,and company operations.Observing to a certain extent,we can find that the stock market holds the economic lifeline of the country's development.How to better predict future price fluctuations in the stock market and obtain stable and effective investment plans has become a research topic for many scholars.This article addresses the shortcomings of choosing a single transaction data source as input in the research of traditional stock market forecasting methods.It takes into account that the stock market is an irrational dynamic system and is also affected by financial news reports in news media and stock-related comments on social platforms.Then,a stock market prediction model that integrates news topics and comment sentiment characteristics analysis is proposed.This improves the forecasting model by adding data sources to predict the future ups and downs of the stock market.First,this article proposes a basic stock market forecast method based on the LSTM model.It inputs historical stock trading data,and then trains 6 basic indicators and 2 calculated technical indicators,and uses the selective memory feature of the LSTM network to realize short-term stock market rise and fall forecasts.Then the model adds text information data on the Internet.It extracts news topic vectors from financial news,and extracts comment emotional features from comments on platforms such as stock bar forums.The text features are added to the input of the stock market prediction model,The experimental results show that the accuracy of the new model has improved and the feasibility of the idea is confirmed.Second,it improves the method of extracting features in the new model.Considering the context of news text,it proposes an attention mechanism and a Bi LSTM model(AB-Bi LSTM)to extract news topic features,highlighting the different importance weights of different news.Considering the actual cost of classification based on semantic rules,the model uses a neural network based on CNN and Bi LSTM models(CNN-Bi LSTM)for sentiment classification and calculation of comment sentiment index.Experiments have proved that under the two optimization methods,the prediction accuracy of the comparison model is further improved.Finally,this article combines two improved methods of feature extraction.It chooses to introduce stock historical trading data,news topic feature vectors in financial news,and emotional features of investor comments in social platforms into the stock market prediction model input.It establishes a stock market prediction model(CAB-LSTM)that integrates analysis of news topics and sentiment characteristics of comments.By comparing the improved model with the previous model and the existing research model,the results show that the prediction accuracy of the CAB-LSTM model is 68.9%,which is better than other stock market prediction models.In practical applications,this can help investors reduce the risk of trading decisions and obtain stable investment returns.