Stock Price Prediction Model For Integration Knowledge Graph And Emotional Analysis

Since the birth of the stock market,the problem of stock prediction has attracted the attention of many researchers from the financial and computer fields.With the development of deep learning,data such as historical stock prices,news,and forums have been applied to stock prediction problems by various deep learning models.However,it still faces many difficulties.For example,financial news data is large and covers a wide range.When the name of a listed company does not appear directly in the financial news,it is difficult to predict the listed companies affected by the news,that is,the mining of related companies in the financial news;There are complex and diverse relationships between listed companies,and it is difficult to model inter-company correlations.In response to the above challenges,this paper constructs a financial knowledge graph to mine companies involved in news texts and the correlation between companies.And based on entity link and knowledge graph embedding technology,a novel stock price trend prediction model based on graph neural network is proposed,which integrates knowledge graph and emotion analysis.Focusing on the issue of stock forecasting,the main work of this article includes:1)Use crawler technology to crawl public data of listed A-share companies from the Tonghuashun website.Then,data cleaning is performed based on domain rules,and Protege is used to construct ontology.Finally,the financial knowledge graph(A-shares Knowledge Graph,AsKG)of domestic A-share listed companies was constructed in the graph database.AsKG currently contains 256,141 entities and 443,874 relationships.2)Designed a strategy to effectively mine financial news-related company information by using the financial knowledge graph AsKG.First,use the improved sequence annotation model BERT-BiDT-CRF to obtain key entities from the news text.Then,based on the entity dictionary and Elasticsearch index,the entity link is performed using the character matching method,and the node corresponding to the key entity in AsKG is obtained.Finally,the node search algorithm based on beam search is used to obtain the corresponding company name with the help of the domain knowledge stored in ASKG.3)Constructed a stock price trend prediction graph with listed companies as the node and inter-company correlation as the edge.The node characteristics are composed of historical stock price data and the results of sentiment analysis algorithms for financial news.The edge feature is represented by the statistical information of historical stock prices and the embedding information of AsKG.Use the graph neural network to analyze the stock price trend forecast graph and predict the stock price trend.Aiming at the above algorithm,this paper carried out a staged experiment,and constructed an experimental data set of financial news data of Tonghuashun website and stock price data of CSI 300.On this data set,the effectiveness of the stock price trend prediction model proposed in this paper is experimentally verified.Experiments show the proposed stock price trend prediction model is effective,and its performance is better than the benchmark model ARIMA.