Friday, May 25, 2007

Artificial Intelligent Software to Classify Online Article Authors

Is it possible to use artificial intelligent software to study authors and writers by their writing? For instance use such tools to find anomalies on Blogs, Internet Forum Posts or searching for human experts by their authored works.

There have been a few interesting programs that can predict is a writer is male or female, but they only work sometimes and not a high percentage of certainty, "Gender Software" and it is far from full proof. The error rate is too high to be viable and such a system can be gamed too easily.

I do not doubt that the project developing such AI software to classify online article authors would be difficult. Yet, I also know it is a lot less difficult once you start and play around with the data a bit, right? And we know it is possible and after working with it for a while, I bet we could improve accuracy, because "We are Writers" with significant attributes, so we are already well within the domain and therefore our ideas and expertise is very much needed if we are to achieve success in this project.

Will we also be able to catch the "bottom feeders" as one top online article writer calls them? He is speaking to the plagiarizers, computer generated article writers and posers. Well I believe we can catch most all of the Plagiarizers now with a little extra work, with very few false positives, although it might happen occasionally.

Computer generated articles, we can catch now, but someday that will be much harder. The Carnegie Mellon crowd already uses there little AI programs to go onto forums and have conversations and can fool most people for a short time period, some people for longer. Key wording bandits, yes they are a problem and with such software they may become very obvious very quickly with such a tool.

No comments: