Improving prediction quality in collaborative filtering based on clustering

Abstract

In this paper we present the recommender systems that use the k-means clustering method in order to solve the problems associated with neighbor selection. The first method is to solve the problem in which customers belong to different clusters due to the distance-based characteristics despite the fact that they are similar customers, by properly converting data before performing clustering. The second method explains the k-prototype algorithm performing clustering by expanding not only the numeric data but also the categorical data. The experimental results show that better prediction quality can be obtained when both methods are used together. © 2008 IEEE.