Abstract: A new recommendation algorithm based on item rating prediction using ants clustering was designed to reduce the sparsity of user rating data.By analysis the web logs of users,ants clustering algorithm was used to cluster users.For the unrated item of target user,a number of nearest neighbor cluster was found,in order to reduce data sparsity,the ratings of other users in the cluster was used to predict the unrated item.Finally,collaborative filtering algorithm was used to provide recommendation services for uses.The web server logs of a self-developed e-commerce web site were used for simulation.The result shows that,the recommendation efficiency and accuracy of our method are both obviously superior to other method.