An Exploration into the Application of Human-in-the-Loop Technologies for Personalised Music Recommendation

Ellis, Christopher (2020) An Exploration into the Application of Human-in-the-Loop Technologies for Personalised Music Recommendation. PhD thesis, University of Nottingham.

[img] PDF (Thesis - as examined) - Repository staff only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Download (7MB)

Abstract

The evolution of the internet over the last 30 years has drastically changed the way we find and consume music. Today we can get near-instantaneous access to vast libraries of music with streaming services like Spotify offering archives in excess of 30 million tracks. Faced with such overwhelming choice it can be easy to become paralysed by the possibilities. We need some automated and effective means of navigating this sea of content to identify the music we want.

The currently accepted solution to this problem is the music recommender system. At their core, modern music recommenders are computer programs which suggest music to users by attempting to accurately predict their taste preferences and identify corresponding appropriate tracks to recommend from a digital musical archive. Unfortunately, in recent years it has been increasingly found that predicting music in this way often produces accurate but obvious, impersonal and uninteresting recommendations that are not necessarily useful or desirable to users. This has lead to the rise of a problem which has become known within the industry as the personalisation problem. In essence, systems are producing recommendations which may be accurate but which are perceived to be impersonal.

In this thesis, we consider how allowing the user to manually engage with and influence the outcome of these automated systems could mitigate this problem and lead to more personal and better-received recommendations. We advocate a human-in-the-loop (HITL) approach to music recommendation that puts the user back in control of their recommendations.

The core contributions of this thesis are:

1. An explanation as to the dangers of solely pursuing predictive accuracy in music recommendation

2. A deconstruction and exposition of the personalisation problem for music recommendation.

3. An evaluation as to the role and significance of considering the intended purpose for which a recommendation is being sought when producing recommendations

4. The development and initial validation of a novel HITL strategy for combating the personalisation problem

Item Type: Thesis (University of Nottingham only) (PhD)
Supervisors: Benford, Steve
Wilson, Max
Kelafidou, Genovefa
Keywords: music, streaming services, Internet, music recommender system, human-in-the-loop, HITL, algorithms
Subjects: Q Science > QA Mathematics
Q Science > QA Mathematics > QA 75 Electronic computers. Computer science
Faculties/Schools: UK Campuses > Faculty of Science > School of Computer Science
Item ID: 61116
Depositing User: Ellis, Christopher
Date Deposited: 08 Feb 2024 14:27
Last Modified: 08 Feb 2024 14:32
URI: https://eprints.nottingham.ac.uk/id/eprint/61116

Actions (Archive Staff Only)

Edit View Edit View