Musical facets, tags, and emotion: Can we agree?

Given the recent explosion in the use of folksonomies to describe digital objects, it is important to study how users tag items in order to design online systems that best meet their needs. This small-scale study explored what music information facets users of the social music website last.fm might consider when assigning emotion-based tags to describe songs, as well as whether a small group of researchers could reach acceptable agreement levels in describing the music information facets present in a small body of music. The top ten songs most frequently tagged with each of five emotions (Happy, Sad, Anger, Disgust, and Fear), as well as synonyms for these emotions, were collected, for a total of fifty songs. Six researchers analyzed the songs’ facets qualitatively. Content analysis on the Pitch and Temporal musical facets revealed that Happy songs elicited the most agreement among the researchers, while Hate songs extracted the least agreement. An assimilation of the other facets and free-text comments exposed general agreement, but significant individual differences were also discovered, making agreement for the purpose of music information retrieval inconclusive. Due to wide variations in the Pitch and Temporal facets of the songs, it is possible that last.fm users may not use these facets as much as the Harmonic, Timbral, Textual, and Bibliographic facets for emotion-based tagging. A similar, but larger-scale, study is planned.