Study On Key Technologies Of Collaborative Filtering Recommendation System Based On Clustering And Neural Network

Due to the advent of the era of big data,the network information resources are exploding,which not only makes it difficult for users to find the information they want but also makes the problem of information overload more serious,which promotes the development of recommendation technology.It is not difficult to see that the recommendation system is one of the effective measures to improve the above problems.To date,the recommended algorithm for collaborative filtering is still one of the most widely used recommendation algorithms.The recommendation algorithm of collaborative filtering mainly finds the neighbor set of the target user by the similarity of the user and then recommends the first n items of the user preference of the neighbor set to the corresponding target user.Although the technology has achieved good application and success in the recommendation system,there are also some problems,mainly cold start problem,data sparseness problem,time factor problem,and recommendation accuracy problem.In view of the above background,this paper is based on the research work of these problems faced by the recommendation system,which is mainly divided into two parts:Firstly,in order to improve the cold start problem,data sparseness problem and time factor of traditional collaborative filtering recommendation algorithm,a user-preference and distance-weighted clustering algorithm combining time factor are proposed,which can make the recommendation result more in line with the user's hobby,thereby increasing user satisfaction.In this paper,the user's basic objective features are introduced to alleviate the user's cold start problem;the improvement of the sparsity problem is mainly by introducing the project type feature,that is,introducing the project feature into the user-item scoring matrix to obtain the user-project attribute of the small dimension.Scoring matrix;project features are also introduced when constructing user-project attribute preference matrix using TF-IDF algorithm,and the influence of user interest drift over time on user preference is considered;weighted post-Euclidean distance is obtained based on the above three matrices Then,the K-Means algorithm is used for clustering,and after a large number of correlation calculations,recommendations are made.Secondly,the recommendation system based on the rating prediction model does not consider the time background and the different categories of the behavior of users.In the past,user behavior analysis mainly considered the user's subjective scoring behavior and did not consider the user's objective behavior,time-based user preference behavior and the behavior of current popular elements.Therefore,in this paper,a new model CNN-BP is proposed to predict the user's next viewing behavior.In this paper,the convolutional neural network CNN is used to analyze the four behavior sequences mentioned above,and the four basic probabilities of each unwatched project are obtained.Then we use the inverse propagationneural network BP to calculate the corresponding final probability and select the first N items with high final probability for the recommendation.The CNN-BP model solves the above problems and the accuracy of recommendations well.Finally,this paper takes the film recommendation as an example and carries out comparative experiments on the MovieLens dataset.The experimental results show that the proposed algorithm has better recommendation quality and performance.