Inspired by a post to the python-list about using spambayes to classify defunct or out of date event listings,(via Uche Ogbuji), and a recent chat with Kendall I started playing with using it to identify spammed or inappropriate listings on Protest.net (and by extension other open publishing web applications). Initial progress was not promising. I think it’s going to be need more time then a distracted afternoon at Coffee Exchange when I was really supposed to be working on a cover letter or revising the resume.
The net simply wasn’t converging (learning) fast enough to get useful results. I think that I need to spend some time thinking about my token choices, which will be substantially different then those used for email, to really get this to work.
Deep down I really feel like these learning, emergent technologies are going to be an important enabler for the new wave of web experimentation going on under the moniker “social software”.