Research On Collaborative Filtering Recommender Algorithm Based On Self-organizing Clustering

With the rapid development of the Internet age,information on the Internet becomes more and more complex and diverse.It's increasingly difficult to obtain useful information on the Internet for its user,which leads to the gradually worse problem of information overload.The emergence of personalized recommendation is,to some degree,a solution to the problem above.Personalized recommendation is done by the recommender systems(RS),and the most important part of recommender systems are recommender algorithms,and one of the most well-known recommender algorithms is collaborative filtering algorithm.Collaborative filtering algorithm often faces data sparsity and flexibility problems.In order to solve the sparsity problem,existing research usually makes clustering before collaborative filtering.However,most of existing collaborative filtering algorithms which makes clustering in advance suffers from poor flexibility because it only supports offline learning and can't be applied to incremental learning scenarios where information is frequently updated.Even a small part of aforesaid algorithm which has better flexibility and can be applied to incremental learning scenarios still exists some deficiencies.For example,the accuracy of recommendation is not good enough and the clustering phase requires a long time to determine the number of clusters.Aiming at the deficiencies mentioned above,this thesis firstly studies the incremental collaborative filtering algorithm which makes clustering in advance,and proposes a collaborative filtering algorithm based on a self-adaptive cluster number clustering algorithm--EEICF(Enhanced Efficient Incremental Collaborative Filtering).Compared to existing algorithms,EEICF can in some degree improve the accuracy of recommendation,and its cluster number will self-adaptive increase during the running time,which effectively reduces the time to determine the number of clusters in the clustering phase.The research work of this thesis are as follows:1)Proposing the MWOSK-means(Modified Weighted Online Spherical K-means)algorithm to improve the accuracy of recommendation by improving the method of calculating user weight and item weight in the clustering phase of existing EICF algorithm.2)Proposing the MGSoC(Modified Growing Self-organizing Clusters)algorithm to realize the self-adaptive growth of the number of clusters during the running time by improving the self-growth process of multiple trees in the existing GSoT(Growing Selforganizing Trees)algorithm.3)Combined with the MWOSK-means algorithm,the MGSoC algorithm and the collaborative filtering algorithm,this thesis proposes the EEICF algorithm which has better accuracy of recommendation,self-adaptive number of clusters,and supports incremental learning scenarios.