A Reputation Framework

Today on the Diane Rehm show, Diane's guests were Jennifer Golbeck, research associate, Institute for Advanced computer Studies, University of Maryland, College Park, Md, Albert-Laszlo Barabasi, professor of physics, University of Notre Dame and author of "Linked: How Everything is Connected to Everything Else", and Kathleen Carley, professor, Computer Science, Institute for Software Research, Carnegie Mellon University. The topic was Social Networks and the Web

At one point Diane said something like "But you don't know who these people are who are contacting you. This is an identity issue!" Indeed. In fact there are two issues.

When Diane says "identity" in this context what she's talking about is a way to link an online persona with meat in the physical world. That is, some way to hold people accountable in the offline world for their acts in the online world. Criminal and civil accountability is a key factor in how we govern society. Courts of law are really just big authentication exercises, if you think about it.

The second issue is reputation, which short of criminal or civil accountability, is the primary way we hold people responsible for their actions. I've been thinking about reputation a lot lately. I believe that the emergence of identifiers which can be used out of context (like OpenID, LID, and i-names) will allow cross-context reputation systems to emerge as well.

Some of my students and I submitted a paper to DIM2006 last week on the reputation framework we're working on. The paper is called A Framework for Building Reputation Systems. Here's the abstract:

This paper introduces a set of principles for governing the design and operation of online reputation systems. We also introduce the design, architecture, and implementation of a flexible, general-purpose framework. called Pythia, for building reputation systems and show an example application of the framework.

If you're interested in reading the paper as a PDF, here's a copy. The paper was produced from the HTML source using Prince XML, so it's identical in content to the Web version.