| Stocks are an important investment tool in the securities market.At present,the total market value of major global stock markets has exceeded US$100 trillion,of which the total market value of Chinese stocks listed globally has exceeded US$18 trillion.Such a huge scale makes more and more investors and scholars participate in the research of the stock market.Correct stock prediction can construct an effective stock investment strategy,help investors reduce risks and obtain better investment returns,so stock prediction has significant research significance.Due to the complexity of the stock market,stock prices are affected by many factors such as policies,economic developments,interest rates,capital flows and investor sentiment,which makes it very challenging to correctly predict stocks.The volatility of stock price is affected by many factors,including other stocks related to it.Stock prediction based on graph learning uses various graph neural networks to learn the interaction between stocks to provide more information for stock prediction.Most of the existing stock price prediction research based on the stock relationship tends to adopt statically defined stock relations based on prior knowledge(such as industry relations and Wiki relations),making it difficult to capture the interplay between stocks over time.In addition,their predictions mostly rely on a single stock relationship,while many types of stock relationships affect the volatility of stock prices in a complex and intertwined manner.In order to capture the dynamic relationship of stocks,this paper combines dynamic time warping(DTW)algorithm and neural network,and builds a new type of price similarity relation graph based on multiview stock price similarity.On the basis of constructing price similarity relation graph,Wiki relation graph and industry relation graph,this paper further proposes a multi-relational graph attention ranking(MGAR)model.The MGAR model comprehensively utilizes the interrelationships between various stocks,and uses an adaptive learning mechanism to realize the effective aggregation of multiple relation graphs,thereby forming an effective relational embedding for prediction.For a given set of stocks,the MGAR model gives a ranked list of future returns,and maximizes investment returns by selecting K stocks with the best returns to trade.This paper has done a lot of experiments on two stock datasets of NASDAQ and NYSE,and the experimental results illustrate the effectiveness of the model in stock ranking prediction.The MGAR model proposed in this paper predicts the ranking list of stock returns in a learning-to-rank manner to maximize the return rate.This paper introduces the relationship between stock prices into the model,so as to mine the dynamic price relationship between stocks such as leading and lagging or rising and falling at the same time.In addition,this paper constructs multiple stock relationship graphs,and uses the graph convolutional network and self-attention mechanism to effectively fuse them,which improves the problem of insufficient effective information caused by the sparsity of a single stock relationship.This provides a new idea for the fusion of multiple stock relationship graphs. |
