User based hybrid algorithms for music recommendation systems

  • Murtadha Sami Luaibi Al-Maliki

Student thesis: Doctoral Thesis


The amount of music available digitally is overwhelmingly increasing. The main purpose of music recommendation systems is to suggest quality relevant songs that fit with the user’s preferences. Currently, most of the streaming music systems recommend songs based on Collaborative Filtering and Content-Based filtering techniques. However these systems fail in dealing with the Cold-Start problem. This thesis presents user-based hybrid algorithms for music recommendation systems to address the Cold-Start problem and to recommend music for both new and existing users based on their context by integrating the social information to provide context aware personalized music recommendation.

This thesis makes two major contributions: First, hybrid recommendation algorithms are developed using multi-strategy approach to give more accurate recommendations by combining collaborative filtering, content based and the user’s context obtained from social network in order to provide both new and existing users with an easy way to discover new songs. In this way, the system is able to estimate what artist/song would match user preferences. Second, a generic Context-Aware Personalised Music (CAPM) framework is proposed for supporting the rapid development of context-aware music recommendation systems and for clarifying the whole process of recommendation. As there are myriad approaches of recommendation, there is a need for a generic framework not only to gather these approaches, but also to interpret them under the proposed framework. Recommendation algorithm types differ by the input structure. For example, social recommendation algorithm uses social information, collaborative filtering uses users rating data, whereas content based recommendation uses item’s characteristics. This difference affects enormously the representation of data and consequently the process of recommendation. CAPM is able to present different input data and uniforms the recommendation process.

The proposed algorithms and the framework have been successfully evaluated via practical experiments by real users. The practical experiments are carried out by presenting a Context-Aware Personalised Music (CAPMusic) application in Google Play which helps users to discover new artists, albums or songs. Satisfactory results have been obtained which indicate that using the proposed hybrid recommendation algorithms leads to better results compared with using the pure content based and collaborative filtering techniques.
Date of AwardSept 2018
Original languageEnglish
SupervisorLinda Yang (Supervisor)

Cite this