Design And Development For Financial Products Recommendations Based On User’s Investment Behavior

With the increasing activity of China’s financial industry and the rapid development of technology,the variety and number of various financial products(such as funds,bonds and stocks,etc.)are gradually increasing,and it is a very important and challenging task for investors to find the products they need quickly and accurately among the many financial products.For securities companies,which have a huge amount of user information and user behavior data,it is urgent to make full use of and tap the value of these business data to uncover user preferences and recommend suitable financial products to users in time,so as to achieve growth in trading volume and profit.Recommendation systems can effectively alleviate the above information overload problem.In this context,this paper conducts a lot of research on the application of recommendation algorithms in the field of financial investment,proposes a recommendation algorithm that combines user sentiment analysis with user investment behavior,researches personalized recommendation algorithms applicable to financial products by analyzing user investment behavior,combining user historical transaction data and user review information,and designs and implements a complete personalized recommendation system for financial products.The specific analysis contents of this paper are as follows,(1)In this paper,we propose AE＿xDeepFFM,a recommendation model based on user’s investment behavior analysis,adding Field-aware Factorization Machine(FFM)to xDeepFM,optimizing and improving the network structure of xDeepFM by using multi-headed attention mechanism,which is able to predict user’s interest in a given financial product based on user’s natural attributes and historical In order to alleviate the cold start problem,the generated recommendation list is reordered based on the user group preference analysis to obtain the final recommendation list of financial products.Through several sets of comparison experiments,it is shown that the AE＿xDeepFFM algorithm proposed in this paper has significantly improved over traditional recommendation algorithms in all algorithm evaluation indexes.(2)In this paper,we propose a deep sentiment analysis model BERT-BiGRU based on user reviews,using the pre-trained model BERT as the teacher model and BiGRU as the student model,and migrating the knowledge learned by the BERT model to BiGRU by knowledge distillation to achieve automatic sentiment classification of stock review texts.Experimental results show that the BERT-BiGRU model proposed in this paper has higher prediction accuracy than models such as BiGRU,while the introduction of BERT model can significantly improve the accuracy of sentiment analysis.The weighted fusion of the obtained user sentiment feature vector and the user features extracted in Chapter 4 is used as the input of AE＿xDeepFM model,which can better explore the real interests of users and achieve high-quality recommendation services.(3)We designed and implemented a financial product recommendation system based on AE＿xDeepFFM and BERT-BiGRU algorithm,introduced the general architecture of the recommendation system and the recommendation process,detailed the implementation principle of the relevant modules of the system,and focused on the processing process of offline analysis and online recommendation.The system has been tested to be functional and stable,and can recommend financial products to different users quickly and efficiently.