Application And Design Of Bank Financial Recommendation Platform Based On Convolutional Netual Network

With the strengthening of residents' financial awareness,the Internet platform and algorithm models are fully used to deal with the work done by traditional financial advisors,such as customer risk preference judgment,asset allocation and other work.This article recommends financial products through online recommendation platforms to improve customers.Satisfaction with financial interactions.On the one hand,with the increase of residents' income level,more funds will enter the wealth management market.On the other hand,the improvement of relevant national financial policies provides security for residents' investment and wealth management.The wealth management market has a wide variety of funds,and the number of funds under each category is large.When residents are faced with a dazzling array of funds,how to effectively choose fund products has become a financial problem for residents.In this paper,by analyzing the advantages and disadvantages of different recommendation algorithm models,a convolutional convolution model based on convolution network is constructed,and a combined product recommendation platform based on the convolutional convolution model is designed and implemented.The data-driven recommendation portfolio is less affected by the investor's own emotions during the financial management process.Based on the data model,the wealth management recommendation platform provides optimized asset portfolio solutions to ordinary customers at the middle and low levels,and performs dynamic risk management on the recommended portfolio,which helps to avoid risks for wealth management customers.The main research contents of this paper are as follows:1.According to the mean-variance model,the funds are divided into equity,cash,fixed-hand and other assets,and the corresponding categories of fund products are selected from the wealth management product pool,and then the wealth management product portfolio is constructed according to different proportions.2.Analyze the problems faced by traditional recommendation algorithms: the cold start problem and data sparse problem of collaborative filtering recommendation algorithms;content-based recommendation algorithms cannot perceive changes in users' interests,and thus cannot maintain efficient recommendations for a long time.This paper proposes a recommendation algorithm based on atrous convolution to solve the problems faced by traditional recommendation algorithms.3.Establish a sequence in the time dimension based on the user's investment behavior,capture the user's transaction behavior data,and deeply mine the interest changes behind the behavior.In this paper,by designing a hole convolution kernel,selecting different numbers of holes to compare the model effects on the three indicators of accuracy,recall rate and AUC,and establishing the hole convolution kernel expansion rate of 2 has the best recommendation effect.Using Vue front-end components and micro-service back-end architecture,a CNN-based recommendation financial management platform was designed and implemented.