Research On Collaborative Filtering Recommendation Algorithm Combined With User Clustering

With the coming of Web 2.0,the repaid development of the Internet makes it easier and easier for users to get information.When faced with a large amount of data,it is hard for the users to find useful information in a fast and accurate manner,which is called information overload problem.Personalized recommendation system is an effective way to tackle information overload problem,its core is recommendation algorithm,and collaborative filtering algorithm is a mature and widely used recommendation algorithm.Every recommendation algorithm is subject to data sparsity,which is manifested as inaccurate calculate of user similarity.To solve the problem of data sparsity,this paper intends to study from the perspectives of user clustering in advance and improving user similarity algorithm,and it proposes SKCA algorithm and RWDS algorithm respectively.Then,combining the two algorithms,this study proposes a collaborative filtering recommendation algorithm combined with user clustering.Starting from the viewpoint of user clustering,the topology potential field theory in physics is brought in and topology potential value is used to show user's importance;then,aiming at overcoming the shortage that users should ensure the number of category by themselves when using K-means cluster algorithm,SKCA algorithm is proposed.SKCA algorithm can ensure the category number adaptively and give the result of clustering and the representative user of each category.All the users need to do is to choose the category of the nearest representative user to go on with collaborative filtering algorithm.Using the dataset of MovieLens,contrastive experiments with other clustering algorithms are carried out,and results show that SKCA algorithm can improve the recommendation quality.Starting from the viewpoint of improving user similarity,higher weights are first given to those ratings,which,according to user's ratings,differ from user's rating habit to realize weighting items.Then,based on Jaccard similarity,which only considers the common rating ratio of two users,the difference of two users' common rating items are taken into consideration to realize similarity improvement;and RWDS is proposed when combining the two methods.RWDS not only considers user rating's global preference but also utilizes rating's professional meaning.Using the dataset of MovieLens,contrastive experiments with other clustering algorithms are carried out,and results show that RWDS can improve the recommendation precision.Finally,a collaborative filtering recommendation algorithm combined with user clustering is proposed when combining SKCA algorithm and RWDS algorithm.Contrastive experiments with traditional collaborative filtering,SKCA and RWDS are carried out separately and the results show that collaborative filtering recommendation algorithm combined with user clustering can remit the influence of data sparsity effectively and improve recommendation quality.