In order to solve the problem of users’ preferences changing frequently

an iterative computing method was presented to gain the weights of users’ preferences as time goes.A bipartite graph was constructed to show the relations of users’ interests and topic classes.On this base

a novel topic-based clustering（TBC） algorithm was proposed to group the nearest neighbors according to users’ interests

which had changed the usual hard partition method meaning "one or the other" for the clustering items.And the partitioned domains of users’ interests based on multiple topics was also es-tablished by the algorithm

which not only fully profiled users’ interests and the relations of topics indirectly reflected in different domains

but also could adaptively track the changes of users’ interests.Experimental results show that TBC method has better declustering outcome of users’ interests than both the traditional K-Means algorithm and FCM method belonged to the soft clustering

and the TBC algorithm also has higher recommendation quality and better efficiency in personalized recommender services.