Research On Recommendation Algorithm Based On Massive Bank Card User Behavior

In recent years, the banking industry presents customization slogan, and competition pressure increased gradually, the rate of return on investment is also an important parameter to be noticed, more diversified products use by the user. Therefore, through the analysis of user behavior data, precise orientation, and through effective recommendation algorithm, make the distance of businesses and users closer. Based on analysis of massive user behavior data and the recommendation algorithm research, it is also a hot issue in the research field.In this paper, we proposes a combination type dimensions of time, space based on improved Item-Based parallelization recommendation algorithm. The main research work of this paper lies in:Based on user history behavior analysis, construct a classification model for the user. The introduction of the long tail effect in economics, directional audience more accurate, more fit the user’s real interest, improve the efficiency of merchants recommended. In order to improve the efficiency, by naive Bias classification algorithm to classify the users, to solve the cold start problem for new users. Time parameter into forgetting curve as a user of consumer behavior, improve the rationality of merchants recommended. Space cluster effect into the merchant, Euclidean distance model based on computing the similarity between merchants, based on the vector space model to calculate the cosine similarity between businesses, both are weighted to get the recommendation algorithm combined type. MapReduce model can finish the training and recommend faster makes user behavior data in the treatment of massive, can be solved effectively. The contrast experiments show that the improved algorithm, in the recommendation accuracy and training efficiency compared with the traditional Item-Based recommendation algorithm are improved.The main results of this algorithm obtained:(1) To accurately classify users into the long tail effect, solve the cold start problem in new users; (2) The introduction of time (forgetting curve), spatial parameters, improve the relevance and the recommendation accuracy, solve the cold start problem of new project; (3) Based on the MapReduce model, the algorithm for parallel reconstruction, improve the operating efficiency of the algorithm.