Abstract: For the defects in the calculation of item similarity of traditional item-based collaborative filtering,this paper proposed an improved collaborative filtering algorithm based on multi-level item similarity.Firstly,the multi-dimensio-nal heuristic methods are used to analyze the similarity of items comprehensively by analyzing user’s behavior records in four aspects,including user collective rating items,user activity,user rating timeliness and user rating.Secondly,based on the four aspects of item similarity,a method for calculating multi-level item similarity is designed.Experimental results show that,compared with the traditional item-based collaborative filtering recommendation algorithm,the algorithm based on multi-level item similarity has higher recommendation accuracy rate and recall rate,and lower MAE va-lue.

Key words: Collaborative filtering,Heuristic methods,Multilevel,Item similarity