Research And Application Of Stock Reviews Sentiment Analysis Based On Hybrid Neural Network

Stock comment sentiment analysis is to automatically analyze the sentiment information of stock reviews through computer technology,and make a classification judgment on the sentiment tendency of stock reviews.Changes in stock market will affect the mood of investors.More and more investors are more inclined to post comments containing emotional tendencies on financial websites,which will adversely affect the stock market.When investors make investment,it is very helpful for investors to obtain useful stock review information,so researching the emotional tendency of stock reviews has practical application significance.Stock reviews are unstructured text data that cannot be directly calculated by a computer.Traditional vector expression models often fail to distinguish between polysemous expressions.The research on sentiment analysis algorithms for stock reviews based on deep learning is mostly based on a single neural network,but the single neural network has its limitations.It will be time-consuming and labor-intensive to manually identify and manipulate the huge amount of stock review information.Therefore,this article focuses on following research on above existing problems:1)Designed a sentiment analysis data set for stock reviews.Because there is no public sentiment data set for stock reviews,due to application requirements,a stock analysis sentiment analysis data set has been designed.First select the financial website of the data source,then write a crawler program to crawl the stock review data,and then preprocess the data to obtain a data set,and finally mark the data.First select the financial website of the data source,then write a crawler program to crawl the stock review data,and then preprocess the data to obtain a data set,and finally mark the data.2)Studied the text representation language model.By studying and comparing the advantages and disadvantages of the language models of TF-IDF,Word2 Vec,BERT,and XLNET,In order to solve the problem of polysemous representation,this article introduces the XLNET language model to the financial field for the first time.XLNET language model is used to transform stock evaluation into vector representation to solve the problem of polysemous representation.3)Proposed a stock reviews sentiment analysis model based on hybrid neural network.aiming at the ability of a single neural network to simultaneously extract the semantic and word order information in the short text of stock reviews,the ability to capture two-way semantic features,and the ability to highlight the importance of key features,this paper proposes a sentiment analysis model for stock reviews based on hybrid neural networks.The model first uses XLNET language model to generate text vectors,secondly uses bidirectional long short-term memory neural network to extract bidirectional semantic and word order information,and then uses convolutional neural network to extract features using 3 different convolution kernels and maximum pooling functions,and finally introduces Attention mechanism is used to identify the sentiment tendency of stock reviews.The experimental results show that the sentiment analysis model based on hybrid neural network is better than the sentiment analysis model of single neural network in accuracy,recall and F1 score.4)Designed and implemented a stock comment sentiment analysis system.In this paper,the proposed sentiment analysis model of stock reviews based on hybrid neural network is applied to actual development and displayed to securities investors in an intuitive form for easy reference,which plays a certain role in assisting investment decisions of securities investors.