Collaborative filtering recommender systems based on k-means multi-clustering

Abstract

Recommender systems support users in searching on the internet by finding the items, which are interesting for them. This items may be information as well as products, in general. They base on personal needs and tastes of users. The most popular are collaborative filtering methods that take into account users’ interactions with the electronic system. Their main challenge is generating on-line recommendations in reasonable time coping with large size of data. Clustering algorithms support recommender systems in increasing time efficiency. Commonly, it involves decreasing of prediction accuracy of final recommendations. This article presents an approach based on multi-clustered data, which prevents the negative consequences, keeping reasonable time efficiency. An input data are clusters of similar items or users, however one of them may belong to more than one cluster. When recommendations are generated, the best cluster for the user or item is selected. The best means that the user or item is the most similar to the center of the cluster. As a result, the final accuracy is not decreased.