Abstract: With the problem of data sparsity and cold-start in the traditional collaborative filtering algorithms, a personalized algorithm integrating item rates and attributes is proposed.When measuring the similarity between items, the algorithm takes into account user ratings and item attributes and adjusts the ratio of them for the final similarity according to the spare situation of system ratings.While predicting the score, the user's preference on item attributes is adapted to represent current user's interest on unrated neighborhood items and produce the final recommendation.Experimental results based on MovieLens data set show that the new algorithm makes neighbor recognition more accurately and improves the system recommended quality significantly.