The collaborative recommendation algorithm based on kNN confirms the number of neighbours subjectively

and is not accurate enough to predict by kNN mean weighting calculating.To address these two problems

the maximum and minimum distance clustering algorithm was introduced and improved to design the heuristic clustering model

the model divided the users allodially without the determination of the category numbers

the neighbours of the target users were the users who were in the same category with the target users;then the category similarity was defined to build the category relation between the unscore and score items of the target user in prediction

and the kNN mean weighting calculating was advanced based on the category similarity.The experiments show that this algorithm improves the degree of accuracy (reducing about 0.035 MAE).