Abstract:

K-means clustering algorithms can effectively reduce dimensions when they are applied to recommendation systems. However, the clustering effect is often dependent on the initial centers. And once the target cluster is selected, the recommendation process is executed only according to the target cluster and has nothing to do with other clusters. To solve these problems, we present a bisecting Kmeans clustering parallel recommendation algorithm based on full binary tree. Firstly, the bisecting K-means clustering algorithm is iterated, and during the iterative process the cluster cohesion level serves as the split threshold to form a full binary tree. Then the active users are classified into k leaf nodes (clusters) using the method of level traversal. Lastly, via the MapReduce framework, the process of recommendation prediction can be parallelized onto the k clusters. Experimental results on the MovieLens show that the proposed algorithm can not only greatly improve the accuracy of the recommendation results but also enhance the system scalability.Key words:  

Key words: full binary tree;K-means;clustering;recommendation algorithm;MapReduce