2:00 pm April 14th, 2008 by wkallander

Talking with Professor Wu today helped to crystallize some of the issues I’m interested in working on either for the class, or expanding and working on for my thesis. I need to come up with the latter soon, as I’m slowly nearing the deadline for the qualification examination, which means I need to get my rear in gear.

My area of interest is primarily focused on the application of machine learning to the problem of “anti-social” networks. Whereas most (online and off-line) social network research has focused on networks of people that readily participate in relationships (and allow or even encourage observation of those relationships), anti-social networks are by their nature covert, and therefore seek to hide or inhibit the observation of relationships for the surreptitious community or group participants. Is it possible to identify them as anomolies in a larger social networking context? Can one divine what lies under the surface of the iceberg - only some of the observations are known, and we seek to discover both the hidden topology and the underlying factors governing that structure.

So, I’d like to use the ECS 289M project as a springboard for that area, either as a survey, or to factor it down into the stepping stones necessary to creep up on the problem. So which is it, breadth first or depth first search? In Professor Davidson’s ECS 289G “Data Mining” class last quarter, I looked at different approaches (graph theoretic, information theoretic, and probabilistic methods) for solving the problem of community discovery in graphs. Generally speaking, that is a problem that has ties to the one above. Another that comes to mind is “Link Prediction” in graphs, a problem that I am interested in looking at, and one that has garnered interest in literature recently [1].

At any rate, the topic is large with many pieces, and has some risks, which gives the larger topic the flavor of a reasonable thesis proposal. Too, it is interdisciplinary, which means I probably need someone on my committee that understands the sociological context (Professor Wu mentioned Dianne [last name unknown] from the UC Davis sociology department as a possible contact - need to follow up on that).

Well, those are the highlights, possibly more details to follow to help smooth the rough areas down.

[1] D. Liben-Nowell, J. Kleinberg. The Link Prediction Problem for Social Networks. Proc. 12th International Conference on Information and Knowledge Management (CIKM), 2003.

This entry was posted on Monday, April 14th, 2008 at 2:00 pm and is filed under ECS289M, Research. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.