Factorization Model Based Application To Mobile B Anking Business Recommendation

With the development of science and technology finance,mobile banking has become an important carrier of the bank's sales channels.For precise marketing of mobile banking,it has become an imperative goal for the bank to develop new profit growth,as well as an important way to expand service channels and develop new business.How to use intelligent technologies such as data mining and machine learning to classify customers scientifically and achieve precise marketing,so as to improve the signing rate of mobile banking,is becoming more and more important for commercial applications.In response to this problem,this thesis carried out the following work:(1)Through offline survey feedback and other means,we collect user data including identity information,transaction data and more than 10 features.After taking care of the data such as discrete feature processing,continuous feature processing,missing value processing and noise elimination,the dataset for mobile banking recommendation are constructed.(2)We treat the problem of mobile banking recommendation as a classification problem,compare and analyze the performance of classical algorithms,such as logical regression,random forest,and factorization machines.It is decided that the field-aware factorization machine(FFM)performs best in feature processing.(3)To deal with the deficient of large amount parameters in FFM,weight sharing matrix is introduced based on Bi-FFM to model the interaction of features and features,features and domains,domains and domains at different levels,and then an improved field-aware factorization machine is proposed as FFM+.On the basis of FFM+,the neural network modeling mechanism is introduced to expand it to Neu FFM,so that it has the capability of high-order nonlinear modeling.Experimental results show that FFM+ and Neu FFM can get a better tradeoff between computational cost and classification performance than traditional FFM.(4)Finally,a demonstration system of customer precision marketing application is designed to apply Neu FFM in the actual mobile banking recommendation,which can support the smooth progress of bank related business.