Research On Financial Recommendation Algorithm Based On Machine Learning

Benefiting from the development of Internet technology and the change of users' behavior and concept,Internet finance has become a new hot spot in economic life.Along with the booming development of Internet finance and the diversification of financial products,investors have the trouble of how to choose suitable financial products.Personalized recommendation is one of the most effective ways to solve this problem.At present,recommendation algorithms based on machine learning are widely used in e-commerce platforms,as well as personalized recommendation of music,video and news.However,it has not been widely studied in the field of financial products.Therefore,it is of great significance for every investor and operator to use machine learning to recommend efficient financial products.In view of the above problems,this paper mainly studies the recommendation algorithm of financial products.The data are from financial products and user behavior information in a financial APP.Financial products are products of existing direct Banks,mainly including financial products issued by Banks and funds(monetary funds and pure debt funds)issued by fund companies.The recommendation is divided into recall and sorting.The recall part mainly adopts two methods.First,according to the life cycle of financial products and the change of users' interest in financial products,the time attenuation function was fitted and added into the collaborative filtering algorithm to make the recommendation results reflect the time effect.Secondly,an improved recommendation algorithm of random walk graph pattern is proposed,in which weights are added to the random walk to reduce the low degree of personalization and the high proportion of unpopular products in the random walk algorithm.The precision and recall of financial product recommendation have been improved to a certain extent.Finally,the reordering phase is added to increase the recall results and reorder them.With the help of machine learning method,the XGBoost model is used to synthesize various factors for sorting.Dig deeper into the user's behavior,a more comprehensive understanding of user preferences.Experiments show that reordering can better meet the needs of users,predict the click probability of users more accurately,and improve the recommendation effect.