| With the continuous development of the social economy,stocks are one of the most important components of the investment market,and at the same time,as an "economic barometer",it is a reaction indicator of the macro economy.Compared with other industries,the stock market has high risks and high returns.Therefore,stock trend forecasting has become a research focus.In the past,the focus of research has always been on phenomenon research.With the development of Internet and big data technology,it is now possible to use technology to predict some development trends in economics.With the development trend of "AI empowerment",research in various industries is centered on deep learning,supported by knowledge graphs and graph databases,and research based on cross-domain research has become an important and popular direction.Among them,stock forecasting is a hot topic,and it is very popular as a cross research direction.Based on the existing technology and algorithms,this paper proposes a forecasting model combining LSTM and SENet.Financial market forecasting model,this model is suitable for companies and individuals who conduct big data analysis,companies that research and develop financial products and solutions,and analysts who conduct quantitative analysis to help them intelligently analyze financial markets and analyze quantitative investment.Provide certain technical support.The main research contents of the paper are as follows:1.This paper aims at the phenomenon of unbalanced and ambiguous information expression ability between the stock information word vector and character-level vector in the input layer of the traditional sequence tagging model(BiLSTM-CRF),and uses the stock information named entity recognition through the fusion of attention mechanism.The model(BiSLTM-Attention-CRF)is used to enhance the stock information expression ability of the input layer of the model and reduce the impact of the imbalance phenomenon on the accuracy of stock information.After verification in this paper,BiSLTM-Attention-CRF has certain feasibility.2.This paper proposes a prediction model combining Long Short-term Memory(LSTM)and Squeeze-and-Excitation Networks(SENet).Based on the LSTM model,the model is integrated into the SENet model,making full use of the output information of the hidden layer at each moment,extracting time series features by the channel attention module,and weighting the output results.Through experimental comparison,it is verified that the proposed LSENet model has better prediction performance,in which the MAE can reach around 3.34.3.This paper proposes a method of stock trend prediction model based on LSENet.The collected data is divided into word vectors by name body recognition and relationship extraction,and knowledge extraction is performed on the entity information,and then the relationship between entities is established.Finally,the completed word vector is established and stored in the Neo4j graph database,and the relationship between entities and entities can be visually queried by using Cypher. |
