Enhancing Item-Based Collaborative Filtering for Music Recommendation System

Abstract

Music recommendation system is a tool designed to help users to find interesting music from huge volumes of digital collections. Content-based filtering method and collaborative filtering method are the most used by researchers in the field of music recommendation. The objective of the paper is to enhance generic item-based CF. In item-based CF, items are grouped into clusters based on their similarity. The item clusters thus formed are used in recommendation process. As the numbers of items in clusters are very large, framing recommendation vector based on only item-based CF might not result in a very successful and accurate recommendation system. This paper proposes a research work to enhance generic item-based CF, by combining with K-nearest neighbor method. Music taste of a user varies based on the time of a day. To include this parameter in generic item-based CF, context information is defined for each item based on the time of the day. Context information is combined with item-based CF to further enhance the proposed system. Proposed methods to enhance the item-based collaborative filtering are experimentally verified by using a standard benchmark dataset Last.fm which is obtained from million song dataset. Results show an improvement over generic item-based CF model.