Recommender System

[leptogenesis]

I guess this could be called site feedback...

Anyway, I'm studying machine learning at Carnegie Mellon University, and for one of my courses this semester (webpage here), I'm supposed to do a data analysis project (see second to last page of the syllabus). I was wondering if, for this project, I could try my hand at creating a recommender system for AMVs using data from A-M-V.org.

All I'd need from the site would be the data about ratings...particularly who gave what star ratings to what AMVs, although other data (e.g. genres or amv artists) might be useful too. In case we're worried about privacy, I wouldn't care if usernames were anonymized, as long as I can tell the different users apart. I think just about everything else of interest is public anyway.

Assuming I manage to create a decent recommender, I'd be happy to help the site admins wire it into this website--I have some experience with php, and a great deal of experience with web programming (mostly Java).

Let me know what you think.

[mirkosp]

http://www.animemusicvideos.org/search/ ... p?method=1

http://www.animemusicvideos.org/search/ ... p?method=2

You mean like those? I'm not sure how they work, but a 3rd method wouldn't harm, I think. It's all about seeing if the admins are fine with it or not.

[leptogenesis]

http://www.animemusicvideos.org/search/ ... p?method=1

http://www.animemusicvideos.org/search/ ... p?method=2

You mean like those? I'm not sure how they work, but a 3rd method wouldn't harm, I think. It's all about seeing if the admins are fine with it or not.

Yeah, that sort of idea. Strange..three years on the org and I never found that feature. I'll have to try it out.

Of course, the assignment also asks for:

Theory. This section should contain a cogent discussion of the theoretical properties of
the method. It should also discuss under what assumptions the methods should work
and under what conditions they will fail.

Meaning I'd need to do something a bit more sophisticated than what those two algorithms claim to do. Of course, whether it will actually work better because of all the fancy mathematics is a question that can only be answered after it's done.