Personalized Recommendation Method Based On User Rating And Category Clustering

With the rapid development of the Internet,people have entered the era of information overload.Users looking for information on the network is like looking for a needle in a haystack,while the recommendation system can actively recommend the information they are interested in.Collaborative filtering algorithm,as the most widely used recommendation algorithm,has been studied by many scholars at home and abroad.Many optimization and improvement methods have been proposed,and the effect is good.However,problems such as sparse data,one-sided attention to user rating information,user interest drift and so on still exist and exist for a long time.This paper proposes a personalized recommendation method based on user rating and category clustering(URCC-CF),which optimizes the traditional user based collaborative filtering algorithm.The main research work includes:(1)The user’s interest preference will change with time,that is to say the problem of user interest drift.In this paper,the Ebbinghaus forgetting curve trend is used to design the time factor to modify the user’s score,so that the early score weight is low,and the recent score weight is high.SVD algorithm is used to reduce the dimension of the user-item rating matrix,which alleviates the sparsity to a certain extent and improves the accuracy of similarity calculation.(2)The traditional user similarity calculation only considers the user’s score,but the user’s score is not the only factor to measure the user’s preference.The user’s preference for item categories also reflects the user’s similarity to a large extent.In this paper,the usercategory rating matrix is constructed by three indicators: project category preference,project proportion and subjective score preference,and the user preference clustering is carried out according to this matrix,which improves the accuracy of user classification.Meanwhile,the comprehensive similarity is generated by combining the similarity of useritem rating and user-category rating,which improves the accuracy of the prediction score.(3)Because of the randomness of the cluster center,the objective function of the clustering algorithm commonly used in collaborative filtering recommendation always falls into the problem of local optimal solution.In this paper,the flower pollination algorithm is used to determine the cluster center,and the fuzzy c-means clustering algorithm is optimized to cluster the users based on the item category preference,which can effectively solve the problem of local optimal solution.At the same time,only the nearest neighbor search is carried out in the cluster,which reduces the search range and effectively improves the computational efficiency and scalability of the recommendation method.Through the test on the Movielens data set,it is proved that the proposed method has good recommendation performance.Firstly,the optimal parameters involved in the algorithm are determined by comparative analysis,and then compare this algorithm with the other three recommendation algorithms to prove that this algorithm is better than other algorithms in accuracy and coverage.Then,the fuzzy c-means clustering and K-means clustering based on flower pollination algorithm optimization proposed in this paper are compared and analyzed.The experimental results show that the clustering algorithm used in this paper can effectively improve the accuracy of user clustering.