Handling User Cold Start Problem In Recommender Systems Using Fuzzy Clustering

Abstract

The use of recommender systems has grown immensely in the recent years because the count of people using internet has grown at an enormous rate. Different websites have been successful in implementing recommender systems. Various techniques like collaborative filtering, content based filtering, knowledge based filtering etc have made recommendations easy and reliable. Yet these techniques face various challenges like cold start problem, scalability and sparsity issues. Cold start problem occurs when there is no sufficient rating information for a new user who enters the system. Thus no recommendations can be made for this user. Different approaches like k-means clustering, hierarchical clustering etc have been used to cluster users so that a new user could get recommendations based on this neighborhood. But k-means clustering fails when dataset size is huge, thus giving lower accuracy. In this thesis, a novel approach is introduced which implements fuzzy c-means clustering algorithm using RStudio to address user cold start problem. A comparison is made between fuzzy c-means clustering approach and the traditional k-means clustering approach based on different sizes of the user dataset. It is proved that as the number of users increase, fuzzy c-means clustering turns out to be a successful and a more accurate technique than k-means approach to generate better quality recommendations for the new user. Thus, solving user cold start problem.