A modest proposal for a distributed music selection system

Raph Levien, 19 Sep 2000

My (unfinished) PhD research topic is on trust metrics. Advogato was implemented largely to provide an experimental testbed for these trust metrics, to see if they'd work in the real world. So far, the answer seems to be "yes".

Since implementing Advogato, I've done a lot of thinking about how these ideas might be adapted for other applications. I've come to the conclusion that one of the most interesting applications, and also one of the simplest to implement, would be a system for choosing music.

I am not going to implement this system, because I am more than busy enough with other projects. This document presents the design, in the hope that someone else will take up the task of implementing it.

Outline

Much of this design should be familiar to those using Advogato, at least from the outside. The new design should preserve many of the desirable features as well, such as resistance to attacks.

The model of user accounts and certificates is the same as the Advogato trust metric. People will create accounts, and then issue certificates for other people who they feel share taste in music.

The central new concept is that of a "recommendation". A recommendation consists of both some identifying information for the song (for example, a Mojo Nation ID), and some subjective metadata about the song.

Flow algorithm