[Search-l] I'm awake too... plus, an idea

[Ed Whittaker]

Not sure about this idea but if you did it there would be no reason to
select only one algorithm necessarily or evolve only one. You could use
*all* your algorithms up to that point and perhaps use the verdicts on each
of them to weight the contribution from each algorithm.

Might end up being computationally expensive but I believe MSoft are doing
something a bit similar for live with about 200 or so different features
including standard "features" like tf/idf, and page quality, all combined in
a neural network. Somehow. I have no details sorry.

-Ed

On 5/1/07, Aerik Sylvan <aerik at thesylvans.com> wrote:
>
> Okay, it's 10:01am in Silicon Valley and I'm awake and even reasonably
> alert :-)
>
> Here's something I don't know if I've mentioned before:  One thing we
> could do to apply the community to improve search results is to constantly
> rotate in different algorithms and let users rate the result relevance.
> Sure, maybe Joe Averageuser wouldn't want to have much to do with that, but
> even a dedicated core of few hundred power users could quickly create
> significant statistics about what algorithms are *actually* working better.
> But you'd never want to stop - constantly take in new ideas about what data
> to collect and how to weight it, then collect data about how well the
> algorithm is working in real time.  Use the feedback from that to "evolve"
> the algorithms.
>
> (Cross-posting this to Brainstorming as well).
>
> Aerik
>
> _______________________________________________
> Search-l mailing list
> Search-l at wikia.com
> http://lists.wikia.com/mailman/listinfo/search-l
> Change options or unsubscribe:
> http://lists.wikia.com/mailman/options/search-l
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.wikia.com/pipermail/search-l/attachments/20070501/e660385c/attachment.html