Stock Trend Prediction Based On Graph Convolution Hybrid Network Model

With the expansion of the financial market,more and more people are buying and selling stocks as an investment method,and the problem of stock prediction has attracted extensive attention.Stock prediction aims at predicting the rise and fall of stock prices,and can assist investors in analyzing and determining trading strategies.Due to the high volatility and non-stationary characteristics of the stock market,accurate trend prediction is very difficult.Therefore,it has brought great challenges to the financial field and computer field.The traditional stock trend prediction methods usually only rely on stock trading data,and ignore the important information implied in the stock news text,which will also change the direction of price change.For this problem,this paper combines with the characteristics of stock price and news text,and proposes a gated recurrent unit prediction model based on multiple features.In order to obtain news text features,this paper adopts a pre-trained Word2 vec word vector model and attention mechanism.Firstly,word2 vec model is used as the word embedding layer to calculate the embedding vector of each word in each news every day,and then the embedding vectors of all words are averaged to obtain the vector representation of each news every day.Because there are many news in everyday,finally,the weight is assigned to each news vector through the attention mechanism,and the weighted average of these news vectors is calculated to obtain the daily news vector representation,which is used to represent the news text features.The difference calculation is carried out on the historical closing price of the stock to obtain the rise and fall range sequence of the stock,which is used to represent the characteristics of the stock price.The two obtained features are input to the gated recurrent unit to capture the timing dependence and obtain the prediction results.Experiments show that this method is superior to the model that only considers the characteristics of stock price as network input.Most of the existing stock trend prediction methods assume that stocks are independent of each other,while ignoring various relationships between stocks.However,the rich relationships between stocks contain more valuable information to realize the prediction.For this problem,on the basis of the prediction model of gated recurrent unit based on feature fusion,the industry relationship between stocks is led into,and a network model integrating graph convolution network and gated recurrent unit is proposed.According to the industry category of each stock,build the stock relationship diagram of the same industry,and learn based on the diagram.The correlation characteristics of the graph are obtained through the graph convolution network,the time dependence of stock price characteristics and news text characteristics is obtained through the gated recurrent unit,thereby completes the stock trend prediction.The experimental results on real data sets show that the model considering stock relationship achieves higher accuracy and improves the prediction effect.