Verifying Music Tag Annotation Via Association Analysis

Music tags provide descriptive and rich information about a music piece, including its genre, artist, emotion, instrument, etc. While many work on automating it, at present, tag annotation is largely a manual process. It often involves judgements and opinions from people of different background and level of musical expertise. Therefore, the resulting tags are usually subjective, ambiguous, and errorprone. To deal with this situation, we seek automatic methods to verify and monitor this process. Furthermore, because multiple tags can annotate each music piece, our task lends itself to multi-label methods which capture the inherent associations among annotations in a given music repository. In this paper, we propose a novel approach to verify the quality of music tag annotations via association analysis. We demonstrate the effectiveness of our approach through a series of simulations using four publicly available music datasets. To our knowledge, our work is among the initial efforts in verifying music tag annotations.