Abstract：It is the key content of traditional collaborative filtering recommendation algorithm to search for neighbor users or neighbor items. In general,data sparsity leads to a decrease in recommendation accuracy. Hybrid collaborative filtering algorithm based on project category preference uses the low-dimensional and binary features of project characteristics for clustering,and uses the category preference information of users to search for nearby users,this kind of method can alleviate the problem of data sparsity to some extent. In order to further improve the similarity between neighboring users,semi-supervised AP clustering algorithm is used to replace the traditional clustering algorithm based on the hybrid collaborative filtering algorithm based on project category preference,and the similarity measurement method is improved. A collaborative filtering algorithm based on semi-supervised AP clustering and improved user similarity is proposed in this paper. This algorithm has two improvements：on the one hand,a new semi-supervised AP clustering algorithm based on k-nearest neighbor density estimation is proposed;on the other hand,Pearson similarity was improved by using user activity factors and user scoring trajectories. Experimental results show the effectiveness of the algorithm.