The Research On Dynamic Multi-Level Collaborative Filtering Recommendation Algorithm

With the rapid development of the Internet,recommender systems has become one of the main methods of information filtering.Collaborative filtering is the most widely used algorithm in the recommender systems,which is mainly divided into user-based collaborative filtering recommendation algorithm and item-based collaborative filtering recommendation algorithm.In terms of accuracy,the existing user-based dynamic multi-level collaborative filtering recommendation algorithm does not consider the influence of time factor on the accuracy of the recommendation result,the weighting function of the existing item-based time weighted collaborative filtering recommendation algorithm is too simple,does not consider the link between the item and so on;In terms of diversity,the existing neighbor diversity algorithm has the problem of low efficiency.In view of these problems,this thesis designs a user-based dynamic multi-level and time-weighted recommendation algorithm,proposes a item-based dynamic multi-level and time-weighted recommendation algorithm,and improves the user-based neighbor diversity algorithm.Specific work as follows:(1)Based on the existing user-based dynamic multi-level recommendation algorithm,the similarity value between users is dynamically adjusted according to the number of same scoring items among users.Meanwhile,a user-based time-weighted recommendation algorithm is proposed,the algorithm according to the score difference between two users to weighted similarity.Experiments show that compared with the existing algorithms,the proposed algorithm reduces the mean absolute error of prediction scores by about 0.25.(2)In the item-based recommendation algorithm,the thesis dynamically adjusts the similarity values according to the number of users who scored two items at the same time,proposes a item-based dynamic multi-level recommendation algorithm,and improves the time weighted function of time-weighted recommender algorithm.Experiments show that compared with the existing algorithms,the proposed algorithm reduces the mean absolute error of prediction scores by about 0.3.(3)In the user-based neighbor diversity algorithm,users who don't score any item are filtered in calculating the relative diversity of the target user and the candidate users.Experimental results show that the proposed algorithm improves the individual diversity and overall diversity by 4.4%and 1.8%respectively,based on the slight change of the mean absolute error and the accuracy of the user-based dynamic multi-level and time-weighted recommendation algorithm.