Question about algorithms

[bannor]

I've been around search engines for quite awhile now and have heard about most of the major algorithms being used in the enterprise search market (PLSA, LSI, Vector, Pattern Matching, NLP etc..).

What i've never been able to figure out is what a comparison of these various algorithms when it comes to precision, recall and relevancy would look like.

Is there a specific technology that stands out if that sort of comparison is done? Or does it truly come down to the user experience and expectations when using the different systems?

Thanks for your time.

Scott

[orion]

Hope this help;

1. You can compare the profile of precision-recall curves
2. You can play with the E-measure.

Orion

PS. We are in La Jolla, San Diego, are you close to us? Perhaps we can meet.

[xan]

Even easier, check out all of the litterature in citeseer or something like that - lots of researchers have written papers comparing algos.

[hardball]

It would depend to a large extent on what ends up in the index. I wonder how many documents in a multi billion index are ever presented to a user or clicked on. Just a guess, but I believe anything beyond a billion or so is probably just noise that distorts any ranking algo.

[bannor]

I will take a look at the websites and tools that were recommended.

Let me know when you want to get together Orion.