Collaborative Filtering Recommendation Algorithm Based On User Context And Item-neighborhood

The vigorous development of Internet technology has led to information overload,making it difficult for the public to discern products and choose products that are most suitable for their needs.Therefore,personalized recommendation system based on information retrieval arises at the historic moment.Collaborative filtering is one of the most widely used and successful recommendation technologies,but it also faces the severe challenge of data sparsity,cold start and so on.Meanwhile,the traditional collaborative filtering algorithms often ignore the influence of the individual context information on the similarity of the item in the similarity calculation.In view of the above problems,in this paper,the constrainted probabilistic matrix factorization algorithm is improved by incorporating item neighbor information,and a collaborative filtering recommendation algorithm based on user context and item-neighborhood is proposed by combining with the user context information and a kind of dynamic prediction and evaluation method.First,the item labels are divided into two categories based on content and emotion,the content and emotion vector of each item are determined by calculating the weight of the two kinds of labels,and the neighbor sets of the items are calculated and integrated into the constrainted probabilistic matrix factorization algorithm;Secondly,according to user context information and the user's emotional settings for the items,select neighboring users and related items;Finally,the dynamic prediction filling method is used to solve the problem of data sparseness in personalized recommendation.On the Movie Lens-1M dataset,the MAE values and Pu values of cosine similarity,pearson similarity and user context algorithm are compared by experiments.The test results show that the algorithm can alleviate the impact of the score sparsity on the algorithm and more effectively predict the user's score on the item,significantly reduce the average absolute error and improve the accuracy of the recommendation.