Application Research Of Quantitative Investment Model Based On Combination Of Deep Learning And Multi-View

The financial sector is one of the important application areas of artificial intelligence.Quantitative investments are currently based on single-view,traditional machine learning and deep learning algorithms.But single-view research has certain limitations.The main contents are as follows:1)Design a short-term timing classification model that combining deep learning and multiview : DMVSTC model.In order to solve the problem of single-view,the model presented in this essay introduces multi-view with deep learning for short-term timing studies.Firstly,five CNN models are trained according to different technical indicators and multiple models are used to identify the stock trend classification,getting the result of image classification view;Secondly,the LSTM model is trained by the stock fundamental data to complete the forecast of stock closing price,using function judge to forecast the result value of the stock price;Thirdly,using web crawler technology crawls the emotion index and uses text analyze the function judges,obtaining the result value of the emotion index view;Finally,the GBDT information fusion algorithm is used to complete the stock timing classification for these three views.In a word,the accuracy of deep learning is more accurate than machine learning as the training sample size increases.The results show that under the same view angle,the multi-technical index is better than the single technical index;in the case of the same technical index,the three views are better than the two views.2)Design DMVSTC model based on sensitive index optimization : SIDMVSTC model.When using GBDT information fusion for traditional machine learning,it is impossible to avoid the issue that special factors determining the results.Therefore,based on the research of DMVSTC model,this essay uses “sensitive index” to limit the value of equation in the information fusion stage,replacing traditional GBDT algorithm with conditional multiple linear regression equation.It is tested by back-testing experiments that the SIDMVSTC model is better than the DMVSTC model.3)Design and Implement Multi-view Data Integration System.In order to obtain data information from three views in the SIDMVSTC model,based on the system's requirements analysis,this essay implements the fundamental data processing sub-system and the emotional index web crawler sub-system.