Exploiting metadata for a song recommender system under scarce data

Authors

DOI:

https://doi.org/10.5335/rbca.v12i2.10047

Keywords:

Filtragem Baseada em Conteúdo; Lista de Recomendação; Metadados; Músicas; Sistema de Recomendação

Abstract

The evolution of digital music made possible, from the 2000s, onwards the emergence of music streaming
services, that produced a highly lucrative effect on the music sector. These services use recommendation
systems to find tracks that will appeals to users, collecting rich information about them and what they listen
to. The obteined volume of data enables a scenario rich in characteristics that describe both, thus making
recommendations more reliable. So, is possible to find recommendations in a scenario with scarcity of data?
The problem this work aims to solve is the difficulty of recommendations systems have to suggesting new
songs for users in front of a scenario with scarcity information, exploring metadata combinations to find
recommendations. Throughout this work a theoretical survey will be conducted, as well as the proposed
system, the necessary models for development this system are presented too. The results show it is possible
to make recommendations in a scenario with scarcity information. The evaluation showed in an unbalanced
environment where only 11% of songs are considered relevant is possible to obtain, in the metric MAP,
approximately 20% of assertiveness along the recommendations lists, as well as a recommendation list with
a larger number of songs helps to obtaining a better result. The evaluation also showed it is possible to obtain
more than 35% assertiveness in the metric MRR, just as the recommendation list with a larger number of
songs interferes directly with the positive metric result.

Downloads

Download data is not yet available.

Published

2020-05-18

Issue

Section

Original Paper

How to Cite

[1]
2020. Exploiting metadata for a song recommender system under scarce data. Brazilian Journal of Applied Computing. 12, 2 (May 2020), 1–15. DOI:https://doi.org/10.5335/rbca.v12i2.10047.