Saturday, November 11, 2006

Musiceuticals: What affects MY mood?

Most of the emails about Musiceuticals so far have fallen into two groups:

  1. I want it, but it seems difficult, so it won't work.
  2. I want it, but everyone is different, so it won't work.
I've been wondering about #2. It's what I referred to in the last post as the question: "What affects MY mood?" In other words, how can we account for the fact that music affects each of us in different ways?

As usual, I don't know for sure. But there are logical starting points that tie back to the two general approaches to song categorization: Algorithms and People. In either case, I think color is a good metaphor. Across cultures, different colors mean very different things. With thanks to About.com, below is an edited list of differences related to death:

Red

  • South Africa: Color of mourning

Yellow

  • Egypt: Color of mourning

Purple

  • Thailand: Color of mourning (widows)

White

  • Japan: White carnation symbolizes death
  • Eastern: Funerals

Black

  • Western: Funerals, bad guys
This variation is only in this one topic. We see similar variation in themes of purity, success, marriage, etc. Who among us on seeing lists like this hasn't found the difference interesting? Are the differences driven by something real and cultural or simply random changes that have taken root. For our wondering, it doesn't matter. Once they take root, they're real. For someone in a particular culture, you can take a pretty good guess that a color will register in their mind in a predictable way. I see Musiceutical customization happening in a similar way. Segments will form with similar reactions. How many segments will there be? Five? Fifty? 50,000? The data will tell us.

And how it might work takes us back to Algorithms vs. People.

If using Algorithms, songs will "cluster" in various ways. Whether by particular cadence, tonality, word choice, etc., songs will cluster. Some will be Musiceutically meaningful. Most won't. This is about people, after all, so people will have to register whether those clusters have meaning for them. Do songs in a particular cluster give you energy? ...bring you pleasure in a melancholy mood? ...get you psyched about an upcoming event? If so, similar songs will bring similar effects. And as people match themselves to clusters, the people themselves start to segment. For example, Group A will find songs in Cluster 23 energizing. Group B will find songs in Cluster 4 sad. Will people who fall into Group A agree on any other song Clusters? It'll be in the data. But I wouldn't be surprised.

If basing everything on People's song ratings, the effect is similar, but you can't really "pre-cluster" songs based on attributes. People's reactions, however, will start to segment with respect to particular songs. Since a pure "People" rated system can't guess in advance what cluster a song fits into, the people themselves have to be rated. That is, a person who consistently rates songs exactly like the rest of their segment becomes a trusted categorizer. If that person says a song is effective at driving or maintaining a particular emotion...it probably does. You can trust that "vote." It's worth noting that this person isn't more a leader or more insightful than anyone else. They just happen to respond more often to songs in a way that represents their segment. For someone who has more random reactions, you don't trust their vote as much and need lots more people to rate a song the same way before it goes in the hopper and gets labeled "an energy song" for those in the segment.

This isn't easy, but it will work and it will happen. The question is which main approach and how it's executed.

No comments:

 
Add to Technorati Favorites