Research And Application Of Collaborative Filtering Recommendation Algorithm Based On BP Neural Network

In recent years,the explosive development of Internet has brought great convenience for us.At the same time,there are many influences,such as information ocean and target information extraction difficulties,etc.Collaborative filtering recommendation is recommended,according to the history of user preference information is one of the effective ways to relieve the information overload,and collaborative filtering recommendation is one of the most widely used technology in the recommendation technology.This essay focuses on the collaborative filtering algorithm of personalized recommendation.By referring to the interests of similar users,collaborative filtering estimates their preferences of certain products and makes recommendations.Aiming to increase accuracy and real timing of this algorithm,improved collaborative filtering algorithms are proposed.(1)Collaborative Filtering with Combined Users and Projects Screening Neighbors,(CUPSN-CF)provides three steps to screen similar neighbors.Before each time,a new screening standard will be introduced based on the previous screening.While searching neighbors,SSN-CF uses the Pearson correlation coefficient method at the first step and considers the influence of users’ attributes at the second step.So users with great characteristic differences can be filtered.CUPSN-CF raises the concept of Prefer set at the last step.Users who have graded the target item or similar items are selected in priority to raise the neighbors’ fitness towards the recommendations.(2)A BP neural network algorithm with adaptive momentum term,which optimizes the problem of slow convergence by improving the BP neural network algorithm.The solution is to change the momentum term to adaptive on the basis of adding the momentum term so that it can change in real time,which well solves the instability caused by the weight adjustment shock.The experiments show that the improved algorithm can greatly improve the efficiency of network training and improve the efficiency of recommendation.According to the experimental results of the algorithm,the improved algorithm for the target user’s neighborhood set and prediction score is more substantial than the traditional algorithm.Recommended by CUPSN-CF algorithm,the recommended results are more suitable for users and improve the user acceptance rate.Compared with traditional scoring methods,the algorithm not only improves the scoring accuracy,but also improves the efficiency greatly.It is shown that the improved algorithm improves the performance of recommendation system.