Study of auto-tagging techniques for domain-specific musical data and considering the long tail problem

• Grant number: 21/15221-9
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: April 01, 2022
• End date: June 30, 2023
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Diego Furtado Silva
• Grantee: Vitor Diniz de Oliveira Cunha
• Host Institution: Centro de Ciências Exatas e de Tecnologia (CCET). Universidade Federal de São Carlos (UFSCAR). São Carlos , SP, Brazil
• Associated scholarship(s): 22/10969-8 - ...And Justice for All Tags: dealing with class imbalance in music auto-tagging using embeddings, BE.EP.IC

Abstract

With the growth of music content streaming platforms, the interest in automating tasks such as recommendation and information retrieval from musical data has been increasing. In this context, social tags have been considered important tools for organizing and retrieving musical data. Tags, or "labels", are data assigned to each music (or artist) to describe it at a high level. Some examples are tags related to genre, musical instruments used and emotions passed by the music. While several machine learning-based algorithms have recently been proposed for the task of auto-tagging, little is known about their suitability to predict underrepresented tags in the training set. The objective of this work is to investigate the performance of auto-tagging techniques for two scenarios: underrepresented tags and songs from different cultures. To do that, a data set containing Brazilian songs will be enhanced through specialized APIs and will be made experiments with artificial neural networks and label propagation.(AU)