[Xapian-devel] About the projects of "Ranking" for GSoC 2012

[Olly Betts]

On Tue, Mar 27, 2012 at 11:19:54PM +0600, Mohiuddin Abdul Qader wrote:
> I have been following Xapian for couple of days. I am very keen on the
> projects of 'Ranking' criteria. "Project: Weighting Schemes"
> is a very interesting project for me, as i have already developed a search
> engine using tf-idf scheme and i would really like to implement tfidf or
> DivergenceFromRandomness on xapian. Will it sufficient to be a GSoC project?

I think just tackling one family or the other wouldn't be enough,
especially since you have prior experience with implementing weighting
schemes (albeit in other software).  You'll also find one much easier
with the experience of doing the other, and looking at performance of
various schemes would be less extra work for additional schemes.

> Another project was very interesting 'Learning to Rank'. I went through
> some study about this project & find out some papers from Microsoft
> Research regarding implementation of learning to Rank using Gradient
> Descent Algorithm (Neural Network). I recently worked on a project on
> Bengali character recognition using a Neural Network. Here the Neural
> Network was build from scratch by me in Java. So it will be exciting to
> work on implementing a neural network & apply learning to rank algorithm to
> see the results.

If there's an existing suitable neural network library (which it seems
there probably is - FANN has been mentioned before) it would be better
to use that rather than put a lot of effort into implementing our own
which we would then have to maintain.

> Can you give any suggestion about what further study I have to do? Also as
> both projects are very interesting which project should i give priority to
> be accepted as a GSoC project?

Parth knows more about this than I do, but there's been some discussion
of the LTR project on the list already, so reading the recent archives
should give you some more useful information.

I'd suggest going for whichever project you find more interesting - we
have had similar interest in both so far this year I think.

Cheers,
    Olly