Verifying tag annotation and performing genre classification in music data via association analysis

Abstract

Music Information Retrieval aims to automate the access to large-volume music data, including browsing, retrieval, storage, etc. The work presented in this thesis tackles two non-trivial problems in the field.

First problem deals with music tags, which provide descriptive and rich information about a music piece, including its genre, artist, emotion, instrument, etc. At present, tag annotation is largely a manual process, which often results in tags that are subjective, ambiguous, and error-prone. We propose a novel approach to verify the quality of tag annotation in a music dataset through association analysis.

Second, we employ association analysis to predict music genres based on features extracted directly from music. We build an association-based classifier, which finds inherent associations between music features and genres.

We demonstrate the effectiveness of our approaches through a series of simulations and experiments using various benchmark music datasets.