Abstract: Aiming at solving the problem of personalized service recommendation in the field of mobile telecommunication network,this paper introduced the context information to the process of personalized recommendation,and proposed a collaborative filtering algorithm based on context similarity and twice clustering.Firstly,according to user context similarity,the users are clustered,and user rating confidence is calculated based on the rating matrix to distinguish core users from non-core users.Secondly,the center of clusters formed by initial clustering can be adjusted according to the rating of core users,and non-core users will be clustered again to form a new cluster.Finally,according to the context similarity,user preferences will be predicted.To some extent,this algorithm can reduce the influence of noise data from rating matrix on clustering results,and reduce the deviation of the recommendation.The experiment based on the simulation data set shows that,the algorithm improves the accuracy of user preferences effectively,and increases the accuracy of collaborative filtering recommendation.

Key words: Recommendation system,Context similarity,Collaborative filtering,Core users,Twice clustering