2007:Previous Discussion on Mood Taxonomy

There are mainly two approaches to set up a music mood taxonomy:

1. Starts from theories in music perception: (Lu, Liu and Zhang 2006) adopted Thayer's two dimensional energy-stress mood model which divides music mood into four clusters:

1. Contentment (low energy, low stress)
2. Depression (low energy, high stress)
3. Exuberance (high energy, low stress)
4. Anxious/Frantic (high energy, high stress)

The authors argue "these four clusters almost cover the basic mood response to music and they are usually in the most highly rated emotions as discovered in [music perception and education]". Also, they pointed out there are alternative adjectives that are equivalent to the above four, and those may be better in describing music mood, such as Tenderness, Sadness, Happiness, and Fear/Anger.

(Feng, Zhuang and Pan 2003) seemed to follow this approach and used four categories, happiness, sadness, anger and fear.

(Li and Ogihara 2003) followed another mood model called Farnsworth model, and gave binary label (existence versus non-existence) based on the ten adjective groups in Farnsworth. We agree with the authors that this many labels made the task too difficult and thus resulted in low performance.

2. Derived from practice of music information services: Popular music websites and software (e.g. AllMusicGuild [[1]]], MoodLogic [[2]]]) seek to exploit emotional aspects of music and provide mood labels for albums or sound tracks. (Mandel, Poliner and Ellis 2006) used mood labels on AMG that included 50 or more songs, which results in 100 mood labels. An advantage of this approach is the ground truth is already provided by those websites. However, such a large number of categories seems overwhelming for the evaluation and post-evaluation analysis. It would be ideal if we could come up a method to cluster those labels into a smaller number of categories (perhaps under the direction of music perception theories). In this way, we can leverage the available labels and keep the contest in a managable scale.

Comments from Cyril Laurier

Looking at the literature, we can conclude that there is no standard in mood taxonomy. Another conclusion is that using many categories gives bad results in the automatic classification, that might be because even humans between them don't have a good agreement. Many categories makes it too overlapped and subjective. But anyway psychological studies have some results about agreement (see Juslin for example). A good solution might be a tradeoff between psychological studies and social context information we can grab from real applications and large online communities (music websites with mood tags from the users).

Comments from Xiao Hu

This is indeed a difficult task, and MIREX is the best place to handle difficult tasks because it gathers the attention and strength of the whole MIR community. As this is the first time for mood classification, I like your idea of making it simpler by setting up fewer mood categories.

Juslin's categories have good musicpsychological roots, but he and Laukka confessed their model took a musican's stand and ignored the socical context of music listening. (Juslin, P. N., & Laukka, P. 2004)

Existing music websites like AMG are right in the social context. So this is their advantage. AMG contains too many categories, so I like MoodLogic more. MoodLogic has only 6 categories, but we are not sure about it without knowing how they came up with the 6 labels. One thing worthy of notice is the categories in AMG and MoodLogic are proposed and organized by experts/editors, not ordinary users.

Link to task proposal: 2007:Audio_Music_Mood_Classification