User context based collaborative filtering recommendation

Abstract

In order to improve the prediction effect of item-based collaborative filtering recommendation algorithm,user context factor was introduced.Firstly the dissimilarity degree matrix of the user context factor was calculated.Then the clustering based on the equivalent dissimilarity degree matrix was adopted to cluster users by dissimilarity value between user and user.After clustering,items that had small dissimilarity value were chosen as neighbors of target item in each user group.These neighbors were used to predict rating of target item for user.Finally,an experiment was given to evaluate the presented approach and to compare it with a typical item-based Slope One algorithm using Movielens dataset.The experimental results suggest that this approach has better performance than Slope One.