Prediction Of Stock Trend Based On Multi-features And Model Fusion

Since stock trend prediction plays a significant role in improving the profit of stock investment,it has always been a hot topic in the academic and financial circles.Investors,researchers and quantitative investment institutions have great interest in stock trend prediction.However,the trend of stock price is the most representative and intuitive reflection in the financial field.The fluctuation of stock price is caused by many factors,including the development level of the real economy,the fluctuation of exchange rate,the market trend,the development level of science and technology,and the mentality of investors.Exploring the relationship between these factors to predict the change of stock prices is one of the problems that need to be solved in the current research.To solve this problem,the effective construction and analysis of the relationship between different characteristics is also a challenge that needs to be solved.In the early days,the stock market could only be traded through telephone,but with the popularization of the Internet,the stock market has developed into a huge trading platform with explosive information volume.Exploring the effective stock prediction performance based on the existing large number of stock prediction sample feature datasets,including historical price information,news information,market technical indicator information,etc.,is an interesting and challenging issue.In the field of computer vision,especially in issues such as image classification,pattern analysis,and image retrieval,Feature fusion has received widespread attention and application as a key method.The core of this theory is to integrate multiple features to have a significant impact on the prediction results,in order to improve the accuracy and interpretability of the prediction.Therefore,in order to address existing challenges,this article proposes a new method,BERTNS,to predict stock market trends related to news information and its underlying emotional implications.In traditional feature fusion methods,the data feature dimensions of news and other text data after word embedding processing are much higher than those of market data features such as historical prices and time series.Therefore,this article uses the pre-trained model BERT and financial sentiment dictionary as news text feature extraction tools.BERTNS can effectively extract both the features of news events and the intrinsic emotional meanings of text content in news articles,And represent it as an emotion vector.Through in-depth exploration of experimental results,this article proposes a stock market prediction method based on model fusion: BERTNSF,which combines various features such as text features with historical prices and technical indicators through model fusion,and adds attention mechanisms to assign weights to different features to predict stock market trends.This article conducted extensive comparative experiments on the main market indices of real domestic stock market data to evaluate the overall effectiveness of the method proposed in this article,and conducted ablation experiments on the proposed model to confirm the effectiveness of each part of the model.The experimental results indicate that the method proposed in this paper improves the accuracy of predicting various stock market trends and to some extent solves the shortcomings of existing research.