Abstract: In many recommendation systems,Collaborative filtering recommendation algorithms based on clustering use some specific algorithms such as the K-means algorithm to cluster the users and items,but the limit is that a user or item can only belong to one category in the clustering result.In practical application,a user may have a variety of in-terests and an item also belongs to multiple categories.To solve the above problem,this paper put forward a novel algorithm based spectral clustering subgroups discovering and C-means clustering,by which we got the user-item subgroups with a high degree of similarities and the membership matrix of subgroups of users and items,which can belong to multiple subgroups.The purpose of our algorithm is to predict the users’ final preference to the items by calculating user’s preference to the items in each subgroup and combining the corresponding membership of users and items in their subgroup,and generate the users’ top-N recommendation results.Experimental results show that our method reduces the data sparseness and improves the recommendation precision and recall compared with previous recommendation algorithms.

Key words: Recommendation system,Collaborative filtering,Spectral clustering,C-means algorithm,Subgroup