Bridging Listeners with Artists: Deep Multi-Objective Multi-Stakeholder Music Recommendations

36 Pages Posted: 20 May 2021 Last revised: 7 Nov 2023

See all articles by Moshe Unger

Moshe Unger

Tel Aviv University - Coller School of Management

Pan Li

New York University (NYU) - Department of Information, Operations, and Management Sciences

Maxime C. Cohen

Desautels Faculty of Management, McGill University

Brian Brost

Spotify

Alexander Tuzhilin

New York University (NYU) - Leonard N. Stern School of Business; New York University (NYU) - Department of Information, Operations, and Management Sciences

Date Written: May 18, 2021

Abstract

Most music recommendation services operate by optimizing the objectives of a single stakeholder (i.e., the listener) based on music listening records. In this paper, we argue that it is more beneficial to consider the objectives of multiple stakeholders (artists and listeners) simultaneously in the utility function for music recommendations, which leads to significant improvements not only in the overall recommendation performance, but also in the welfare of artists and listeners separately. To achieve this goal, we propose a novel multi-objective multi-stakeholder recommendation framework, where we utilize state-of-the-art deep learning techniques to formulate multiple ``tower'' neural networks, where each tower learns one particular stakeholder objective based on the music listening records. These towers are then connected through a novel ``bridge'' architecture between hidden layers of tower neural networks, which bidirectionally share the latent information across the towers to capture the heterogeneous relationships between the different objectives. We further propose a novel Shapley-oriented mechanism to identify the optimal location of the bridge to maximize its information-sharing effectiveness. Finally, we aggregate these predicted objectives through a hyperparameter optimization approach to generate the utility values of candidate music and produce recommendations. We empirically demonstrate the benefits of our proposed framework through extensive experiments on two proprietary industrial-scale datasets provided by Spotify and on one public music listening history dataset from Last.fm, where our model achieves significant performance improvements for both stakeholders (artists and listeners), when compared to state-of-the-art music recommendation models and existing multi-objective multi-stakeholder methods.

Keywords: recommendation, neural networks, multi-objective, multi-stakeholder, transfer learning

Suggested Citation

Unger, Moshe and Li, Pan and Cohen, Maxime C. and Brost, Brian and Tuzhilin, Alexander, Bridging Listeners with Artists: Deep Multi-Objective Multi-Stakeholder Music Recommendations (May 18, 2021). NYU Stern School of Business Forthcoming, Available at SSRN: https://ssrn.com/abstract= or http://dx.doi.org/10.2139/ssrn.3848670

Moshe Unger

Tel Aviv University - Coller School of Management ( email )

Tel Aviv
Israel

Pan Li

New York University (NYU) - Department of Information, Operations, and Management Sciences ( email )

44 West Fourth Street
New York, NY 10012
United States

Maxime C. Cohen (Contact Author)

Desautels Faculty of Management, McGill University ( email )

1001 Sherbrooke St. W
Montreal, Quebec H3A 1G5
Canada

Brian Brost

Spotify ( email )

150 Greenwich St
New York, NY 10007
United States

Alexander Tuzhilin

New York University (NYU) - Leonard N. Stern School of Business ( email )

44 West 4th Street
Suite 9-160
New York, NY NY 10012
United States

New York University (NYU) - Department of Information, Operations, and Management Sciences

44 West Fourth Street
New York, NY 10012
United States

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
331
Abstract Views
1,191
Rank
158,509
PlumX Metrics