Research On Recommendation Algorithm Based On Collaborative Filtering And Partition Clustering

With the rapid development of internet technology and communicationtechnology, people have entered information overload age from informationstarvation age. Currently, information retrieval and information filtering are the mainmethod to solve the information overload, information retrieval technology isbecoming more and more mature. As a supplement to the search engine, personalizedrecommendation has been widely applied to the Electronic Commerce, SocialNetworks, Location-based Services and Personalized advertising, etc. Commonpersonalized recommendation technology contains collaborative filtering,content-based recommendation and graph-based recommendation, etc. Collaborativefiltering is the oldest and also the most successful algorithm among them. Assumingthat the past similar users may have similar behaviors in future, with the help ofanalyzing and calculating the history of the user community behavior, it could findsimilar users and then recommend for a target user. However, when faced with vastamounts of information, Collaborative filtering algorithm is also faced with datasparse, cold start, recommendation accuracy and scalability issues.Recommendation results of traditional user-based and item-based collaborativefiltering are limited to the analysis the historical user-ratings, not consider thecontent attributes of user and item. So mining valuable information as more aspossible is crucial for improving the accuracy of recommendation. In this paper, byutilizing the items’ attribute information, we improve the collaborative filteringneighbor searching accuracy and neighbor searching efficiency, and researches aredone as follows:Firstly, by analyzing a practical example, this paper finds out the deficiency inneighbor searching method of traditional collaborative filtering algorithm，data high dimensional sparseness leads to similarity calculation inaccurate. Then propose aneighbor searching algorithm based on similar rating behavior.Secondly, combine partition clustering and collaborative filtering, the clusteringprocess is to group a set of objects into some clusters so that objects within a clusterhave high similarity, through clustering, reduce the search range for searchingneighbor effectively. Based on this, this paper proposes user-rate clusteringcollaborative filtering, user-interest clustering collaborative filtering and user-ratecombine user-interest clustering collaborative filtering, and then research theirrecommendation real-time and recommendation accuracy.Finally, we design experiment on dataset MovieLens, compared with thetraditional user-based collaborative filtering algorithm, it is proved that our proposedneighbor searching algorithm based on similar rating behavior improve therecommendation accuracy, combine partition clustering and collaborative filteringcould help to improve recommendation real-time and recommendation accuracy.