Abstract: In electronic commerce and information system,collaborative filtering is a very important technique．User similarity measure of the scientific method is crucial.In order to obtain better accuracy,numbers of common Ratings between users were used to dynamically adjust the original similarity to more accuratelly reflect the authenticity of the similarity between users．On this basis,according to the social network FTL model (follow the leader) thoughts,for new users or users who cannot find the nearest neighbor,prediction algorithm based on expert trust degree was used instead of similarity to predict the user’s score,making up the deficiency of the traditional algorithm．Experiments show that the algorithm can improve the prediction score,the accuracy and the quality of recommendation,and alleviate the cold-start problem for new users.

Key words: Collaborative filtering,Recommendation algorithm,Similarity,Cold start