Research On Personalized Stock Recommendation Method Based On Investor Portrait

With the rapid development of China’s financial market and the great progress of artificial intelligence,machine learning and big data technology,the operation and profit model of many traditional businesses in the financial market have undergone subversive changes.Investors’financial needs to meet their asset preservation and appreciation are also rising,and the personalized demand to realize differentiated stock investment trading risk and return preference is becoming increasingly obvious.The continuous rise of intelligent investment advisers has provided opportunities for investors’ personalized financial needs.As an intelligent investment advisory service in the asset management industry,intelligent investment advisers can intelligently evaluate investors’ risk preferences and obtain investment combination suggestions matching investors based on algorithm models,providing investors with personalized choice space for purchasing stocks,Personalized recommendation technology to improve the learning efficiency and accuracy of intelligent investment advisers is the top priority of intelligent investment advisers.As the preferred algorithm for intelligent investment advisers,collaborative filtering recommendation faces problems such as data sparsity,cold start,and difficulty in recommending long tail items,resulting in low recommendation performance and poor user experience.Therefore,it is of great theoretical value and practical significance to crack the personalized recommendation methods and technologies of stocks in intelligent investment advisers and realize the accurate financial services of stock investors.In view of the above problems,the paper combines the user portrait technology and focuses on the basic problem of "key recommendation algorithm for the personalized needs of stock investors".First,the paper constructs the user portrait of the intelligent investment adviser of stock investors,designs three stages of the user portrait construction process,constructs the index system of the user portrait label through the collection of portrait indicators,and calculates the index weight using AHP,Xgboost algorithm is used to build the user classification model of investors.Secondly,from the perspective of association rules,stock content and deep collaborative filtering,personalized recommendation sub algorithms under three scenarios are constructed.Association rules are used to analyze the linkage effects between stocks and between industries,identify the related indicators of stock fluctuations,and extract the stock characteristics from the unstructured information of the sector,basic information and comment information based on stock content,Based on the deep collaborative filtering,the user’s preference intensity of different features is identified,and the multi-stage stock matching is carried out by combining the fuzzy clustering results.Finally,the stock hybrid recommendation algorithm is designed to realize the logical framework of data preprocessing layer,sub recommendation algorithm layer,recommendation algorithm fusion layer and model effect evaluation layer.The multi-stage fusion of the early,middle and late stages of the algorithm is adopted,and an example is used to verify the effectiveness of the model algorithm.The innovation of this paper is mainly reflected in the following three aspects:Firstly,based on the perspective of investor behavior preference,a user portrait model of intelligent investment adviser for stock investors is constructed.Although the existing studies have identified a large number of stock investors’ purchase preferences,they still stay in the traditional portfolio recommendations,while the research on user portraits focuses on the network behavior of Internet users.This research integrates the user investment preference in the stock field with the user portrait technology.Based on the logical framework of the stock based intelligent investment adviser,the user portrait process of data collection,data mining,filtering,label extraction and reorganization is constructed.The portrait label system of stock investment users is constructed from five aspects:investment ability label,behavior feature label,industry preference label,regional preference label and risk preference label.On this basis,based on the algorithm advantages of Gradient Boosting,TOPSIS and FND-LDA2vec,this paper constructs the investor user classification tag model,evaluation tag model and stock bar topic preference mining model,and verifies the effectiveness of the model through numerical example analysis.Secondly,based on association rules,stock content and in-depth collaborative filtering,the sub algorithms of investor stock intelligent recommendation are constructed respectively.Although the existing research has identified the advantages of association rules,content-based and collaborative filtering recommendation methods and technologies,the algorithm has some problems,such as data sparsity,cold start,long tail items and so on.In this study,the three algorithms are improved respectively.Aiming at the association rules,we use their information linkage advantages to mine the industry linkage and individual stock fluctuation trend of the internal association of the stock industry and the stock index,and take into account the two levels of information to realize the stock recommendation based on Apriori;Aiming at the unstructured advantage of content-based recommendation,the stock text modeling methods of TF-IDF and word2vec are used to calculate the similarity between stock text content vectors;Aiming at the problem of data sparsity in collaborative filtering,a personalized stock recommendation list is generated by combining the fuzzy clustering and multi-stage matching of stock pool with the depth learning algorithm to optimize the nearest neighbor collaborative filtering algorithm.Finally,the fusion recommendation system of data preprocessing layer,sub recommendation algorithm layer,recommendation algorithm fusion layer and model effect evaluation layer is constructed,as well as the multi-stage fusion algorithms in the early,middle and late stages of the algorithm.Although existing researches have constructed personalized hybrid recommendation algorithms from multiple perspectives,they have not yet been applied to the field of stock recommendation of intelligent investment advisers,and most recommendation algorithms are based on results,ignoring the compatibility of process and system.This research is based on the overall needs of stock configuration selection and recommendation of stock investors’ intelligent investment advisers.The recommendation sub algorithm based on association rules,the content-based recommendation sub algorithm and the deep collaborative filtering recommendation sub algorithm build a multi-layer fusion recommendation framework system of data preprocessing.The early fusion of recommendation algorithm adopts hierarchical hybrid recommendation technology,the mid fusion of recommendation algorithm adopts waterfall hybrid method,and the late fusion of recommendation algorithm adopts weighted hybrid recommendation technology.The effectiveness of the algorithm is verified by an example.