Music Mesh

Via Paul Lamere’s blog, a new music exploration tool recently came on line called MusicMesh. This application immediately caught my attention because it so closely resembles some of the novel features of Orpheus, and some of the ‘future steps‘ we wanted to incorporate. The graph based browsing, where nodes are artists or albums and the vertices represent finely grained measures of similarity, the rich information panes populated with track listings, discographies, interviews and album reviews, and the use of disparate web-based data, are all components of the original Orpheus application written in 2005. MusicMesh does this style of music exploration well.

MusicMesh also tackles some of the ‘future steps’ we imagined but never achieved in Orpheus. It is web-based, forgoing a fat client altogether; the graph is more visually appealing as it incorporates album art and artist photos directly into the graph; and it allows to users listen to full songs as they browse. In a cool twist, it actually uses YouTube videos to play the music and show some video footage while users listen. This is awesome. In general, the application makes good use of sources that were not yet available when Orpheus was developed, and makes better use of some of the sources that were available in 2004/2005.

What seems to be lacking is any personalized understanding of a users’ tastes. In Orpheus, we knew what music users had in their libraries, and we captured user rating information for that music as well. This data is potentially powerful as one begins to serve up implicit or explicit recommendations in the graph or elsewhere. That doesn’t seem to exist (yet) in MusicMesh.

The meaning of the vertices in the MusicMesh graph are also curious. The majority of first degree neighbors are other albums in an artists discography; and then there are a few random artists thrown in. It’s not immediately clear why these additional artists are presented. What is the connection supposed to represent? It is implied that the data is derived from Last.FM, but the semantic value of the connection remains unclear. Also, the value of displaying the artists’ discography in the graph is questionable. Why not show that in one of the information panes, freeing up graph space for other similar artists or recommendations?

We discovered in our Orpheus user studies that users liked being able to control what measures of similarity were used to construct connections between artists. Some users cared about collaboration, for example, while others were more concerned with label affiliation or sound similiarity. Giving users the power to choose what constitutes the vertices may improve the relevance of these connections, and by extension the browsing experience.

It’s interesting to see how they are dealing with the intellectual property issues that come with using potentially copyrighted data, including music, album art, 3rd party text, and video. Their ‘About’ section is mostly concerned with disclaimers and legalese that suggests you’ll be tarred and feathered if you do what they do: namely use someone else’s content on your site. They claim to have licensed their data from Amazon and YouTube, but I kind of doubt they actually license the music and video content. If they have, I’m sure there are a lot of companies that would love to learn how to convince big media companies to bend over and give it up so easily.

At any rate, it’s pretty cool to see people developing applications like this, and it’s fun to be able to browse around the graph and listen to the music at the same time, and see some video footage of the artist playing. Nice work!