Abstract
Recommendation systems have become popular in recent years. A key challenge in such systems is how to effectively characterize new users' tastes - an issue that is generally known as the cold-start problem. New users judge the system by the ability to immediately provide them with what they consider interesting. A general method for solving the cold-start problem is to elicit information about new users by having them provide answers to interview questions. In this paper, we present Matrix Factorization K-Means (MFK), a novel method to solve the problem of interview question construction. MFK first learns the latent features of the user and the item through observed rating data and then determines the best interview questions based on the clusters of latent features. We can determine similar groups of users after obtaining the responses to the interview questions. Such recommendation systems can indicate new users' tastes according to their responses to the interview questions. In our experiments, we evaluate our methods using a public dataset for recommendations. The results show that our method leads to better performance than other baselines. © 2014 Springer International Publishing.