The SOM Based Improved K-Means Clustering Collaborative Filtering Algorithm in TV Recommendation System

This paper aims on collaborative filtering (CF) in TV recommendation system which combines content-based and collaborative filtering recommendation mechanism, we propose an algorithm that using the self-organizing mapping (SOM) to optimize the improved k-means (IK) clustering in collaborative filtering. The whole clustering algorithm is divided into two phases: at the first stage, the quantity of the preliminary clustering and the central point of each cluster were acquired by means of the auto-clustering advantages of SOM algorithm, at the second stage, we first improve the basic k-means algorithm. We utilize the adjusted cosine similarity to calculate the distance from the user to the cluster center, and when calculating the mean of the cluster, just consider all of those users who have given scores for items. The improved k-means improves the accuracy of clustering and is more suitable for using in CF system compared to the basic k-means algorithm. Furthermore we take the results of SOM as the initial input of the IK algorithm to make the further clustering, the accurate clustering results are gained. Finally, the simulations verify that the MAE (mean absolute error)was reduced by 15.7% and 17.4% respectively compared to IK and k-means algorithms, the proposed approach increases the quality of clustering, and enhances the accuracy of the recommended TV program.