Re: RFC: Presentation on Machine Learning with Perl

[bibliophile]

in reply to RFC: Presentation on Machine Learning with Perl

Ok... this isn't *directly* relevent to your presentation, but it did twig a thought...

I read a lot of online newspapers, subscribe to a lot (too many!) RSS feeds, and have a huge list of sites I try to keep up with.

In my perfect world, I'd have a system that could do a content / context scan of all this raw data, and present me with just the stuff I'm particularly interested in.

I'd write the Parse::MeaningFromText and Mind::Read::MyInterests, but (what with all the reading I'm doing) I just don't have the time.... :-)

[lin0]

Hi bibliophile,

It is a very good thought, indeed. However, you would need to think carefully and extensively on what kind of features the articles you are interested in have in common. You could use some sort of data clustering (FCM, maybe?) to help you with this task. You would then need to find a way to extract those features consistently. Finally, you could use a classifier to filter the raw data and present you only with the stuff you are interested in. When you design the classifier, try to incorporate a confidence index that tells you how reliable the results are. In this way, you could play with the outputs until you are happy with the results. Does it make sense?

Cheers,

lin0

[bibliophile]

It does make sense... at least as far as my (quite limited) knowledge of ML goes :-)

One of my always-backburnered thoughts was to build a neural-net-backed "observer" that would watch my browsing habits for a few months, noting things like how long I spend on a particular page, whether I follow links from it, etc., and from that be able to make predictions on stuff I might be interested in.

One of these days^H^H^H^Hyears....