The Research Of Stock Price Prediction Based On News Sentiment Word Embedding

Both traditional finance and modern behavioral finance believe that stock market fluctuations are affected by information release,dissemination,and public acceptance.Earlier research on stock price prediction mainly analyzed and processed historical stock trading data,from which the features that are beneficial to the judgment of stock price trends were extracted.With the development of technologies such as AI and NLP technology,extracting features from the news media that reflect macroeconomic fundamentals and information affecting investor sentiment became possible,and researchers began combining news text data with stock trading data to predict stock prices.One of the keys to using the news for stock price prediction is to accurately extract the information contained in the news.At present,the mainstream news information representation methods mainly include using the BOW method to represent news as a real-valued vector and the word vector to represent news as a text matrix.However,there are obvious shortcomings in the process of using these two methods to characterize news information.On the one hand,the BOW method is easy to cause information loss,and the ability to accurately express information is insufficient.It can only capture limited information in the text.The limitation of the word vector model itself makes it insufficient to capture the sentiment information in the news,so it cannot fully and accurately represent the news content.Inspired by GloVe,in order to more accurately represent news information,we introduced additional sentiment information in the word vector space,and proposed a new word vector model-Senti-GloVe.This method takes both semantic and sentiment information in news text into account,better characterizing news and extract news features.On the basis of the correct characterization of news information,how to accurately quantify the impact of market information on stock market fluctuations has become the focus of stock price prediction research.In recent years,deep learning methods have been widely used in stock price prediction.Different deep learning models have different characteristics,such as CNN are good at capturing local features in data;LSTM are suitable for processing time series data;self-attention models can better establish dependencies between input sequences.In order to give full play to the advantages of various deep learning models and improve model performance,this paper proposes to use ensemble learning ideas to combine various types of deep learning models to build an end-to-end deep ensemble stock price prediction model;at the same time,use transferring training scheme to improve the stock price prediction performance for base models and ensemble model.The main contributions of this paper are as follows:(1)A new word vector model considering news sentiment,Senti-GloVe,is proposed.With it,we can better characterize news and extract news semantic and sentiment features,which are conducive to stock price prediction.In addition,Senti-GloVe word vectors can also help improve the performance of sentiment analysis tasks;(2)Design a stock price prediction model architecture,use feature engineering to process stock transaction data,and use different deep learning models to model the transaction data characteristics,characterize the news by different word vectors and get the news features,combine the two to make predictions on stock price trends,and verify the positive effect of the Senti-GloVe word embedding on stock price prediction;(3)Construct An end-to-end deep ensemble stock price prediction model that fuses the transaction data features obtained from different deep learning models to improve the generalization ability of the model.Through the training scheme of transferring,the ensemble model guides the base model learning during the training process,further improve the performance of stock price prediction models.This article has made active explorations in the areas of news feature extraction and stock price prediction.It provides a certain reference for researchers in related research fields and has certain practical significance.