Research On Ensemble Clustering Algorithm And Its Application In Personalized Recommendation

Clustering is a key technique in data learning,which is an unsupervised classification method.Generally speaking,clustering is to divide the data into different clusters,and the similarity of data in the same cluster is as large as possible,but not in the same cluster as small as possible.In recent years,clustering has appeared in many new technical research areas,such as Personalized Recommendations.Personalized Rcommendations recommend the information according to the user's historical data and preference habits to the user,mining the potential needs of the user,which greatly reduces the time for finding the information of interest and improves the efficiency of the network platform.However,when Collaborative Filtering Recommendation algorithm faces huge amount of data,the algorithm recommendation efficiency will be reduced.Using the characteristics of clustering algorithm data classification to solve the disadvantages in recommendation can not only reduce the amount of computation,but also improve the efficiency of recommendation.When clustering algorithm is applied in personalized recommendation technology,how to achieve fast and efficient recommendation is a difficult point in research.This paper analyzes the shortcomings of the classical clustering algorithm itself and the problems and shortcomings of the recommendation algorithm,and proposes innovation points as follows:Firstly,aiming at the shortcomings that random selection of the initial center of the K-means algorithm has great influence on the clustering results and easy to fall into the local optimum,the F-KMs clustering algorithm that optimizes the initial center of the K-means by density peaks is first proposed,and then the N-FK integrated algorithm: not only can get the best initial center quickly,but also use the feature of Spectral Clustering algorithm solves the problem that F-KMs cannot process data of arbitrary density and shape.Secondly,in dealing with large-scale data,using AP algorithm alone has high complexity and needs huge memory to support.The algorithm results are greatly affected by parameter values and the AP cannot handle Unconvex data.This paper combines the AP and N-FK algorithms to propose a three-stage “multi-level” integrated clustering algorithm.The first layer uses the AP to classify the data sparse and coarse,each class selects a class representative element;the second layer uses the N-FK to cluster the obtained class representative elements in detail;finally,the results of the first two layers are combined to obtain the accurate division of all the data.The proposed integration algorithm,which combines the advantages of AP and N-FK algorithms,can deal with different types of large-scale data and reduce storage space utilization.Thirdly,this paper applies ensemble clustering algorithm to personalized recommendation,and proposes a multi-level recommendation algorithm based on grass-roots clustering.this algorithm first cluster the original big data set and then rebuild the user matrix from the target class with high similarity for collaborative filtering recommendation,which reduces the computational complexity.By using the MovieLens dataset for experiments and tests with recommended performance metrics to prove that the multi-level recommendation algorithm proposed in this paper improves the recommendation efficiency,and can deal with the disadvantages of data sparsity,make the recommendation more personalized.