A Collaborative Filtering Algorithm Based On Clustering And Slope One

The development of computer technology has brought great convenience to us.However,it makes the information more and more rampant.How to help users to find needed information and reduce the cost of accessing to information becomes a hotspot.Recommendation system which could fully exploit the user's preferences to help the user to make personalized recommendations,so it is widely used.Collaborative filtering is the most mature,the most widely used recommended technology,particularly,it is the most successful techniques in E-commerce System.Collaborative filtering technology makes use of user's evaluation information to give recommendations.However,with the increase of users and commodities,it suffered from data sparsity and scalability problems which lead to inaccuracy in recommendation.This is an urgent problem which should be solved currently.In this paper,we described the collaborative filtering in detail and analyzed the current main recommendation algorithms.We focus on the two problems and propose a collaborative filtering algorithm which combined Slope One and user clustering.For the problem of data sparsity,the method of filling could be valid.The Slope One algorithm is an effective algorithm in solving the sparsity problem.It will reduce the sparsity of rating matrix and improve the recommendation accuracy.The original Slope One algorithm was just in the light of rating data to make prediction and leave out consideration of item attributes similarity,which made the prediction lack of correlation.So we introduce the item attributes similarity to make up the defect.Before we use the Slope One algorithm,we must find the target item's neighbors according to attributes and rating similarity.Then using the Slope One module to make prediction.In this way,we just use partial of rated items and the items we used are related.The prediction score will be more reasonable and explicable.The experiment result shows that our Slope One is more effective in recommendation system.For the problem of scalability,we introduce the idea of user clustering.In line with user's attributes the Ant Colony Algorithm works out the clusters.With the clustering results,the target user's neighbor users will be searched in the same cluster.It will shrink the searching space and enhance the scalability in the meantime.The thesis is validated by the MovieLens dataset.Firstly,we verify the validity of the Slope One algorithm combine with item attributes similarity,then make sure user clustering module works efficiently in recommendation systems,finally,combine the two module and apply to recommendation algorithm to prove the improvement in recommendation accuracy and system scalability.The experiment result shows that the algorithm we proposed could enhance the recommendation accuracy and system scalability to some degree.