Complex multi-value boosting

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Complex multi-value boosting

Michael Lugassy
Guys --

What schema will you use for 500K docs with a variety of 0-30
different category ids, each carrying its own weight and completely
overriding the default scoring?

For example, these documents:
A: 1:0.21, 2:0.41, 3:0.15 ...
B: 1:0.18, 2:0.65 4:0.98 ...
C: 6:0.75 ...
D: 2:0.14 ...

When searching "1" I'd like document A to appear first (has 0.21) and
when searching "1 || 2" i'd like document B to appear first (has an
aggregate score of 0.83 vs. 0.62). Currently I run this with full-text
after artificially repeating the number of each category's weight
(i.e. "1" would appear 21 times on a text field) - is there a better
way?

Best,

-- Michael
Reply | Threaded
Open this post in threaded view
|

Re: Complex multi-value boosting

jrduncans
On Mon, Nov 23, 2009 at 3:39 PM, Michael Lugassy <[hidden email]> wrote:

> Guys --
>
> What schema will you use for 500K docs with a variety of 0-30
> different category ids, each carrying its own weight and completely
> overriding the default scoring?
>
> For example, these documents:
> A: 1:0.21, 2:0.41, 3:0.15 ...
> B: 1:0.18, 2:0.65 4:0.98 ...
> C: 6:0.75 ...
> D: 2:0.14 ...
>
> When searching "1" I'd like document A to appear first (has 0.21) and
> when searching "1 || 2" i'd like document B to appear first (has an
> aggregate score of 0.83 vs. 0.62). Currently I run this with full-text
> after artificially repeating the number of each category's weight
> (i.e. "1" would appear 21 times on a text field) - is there a better
> way?
>
> Best,
>
> -- Michael
>

It sounds to me like you want to use payloads (the same issue I had
recently):
http://old.nabble.com/Customizing-Field-Score-%28Multivalued-Field%29-tp26182254p26182254.html

That thread has some details on the eventual implementation I chose.  Let me
know if you have any questions.  Note that I did use the scoring as a boost,
not "completely overriding the default scoring", but I think the impact is
basically the same, as was satisfied it was good enough.

--
Stephen Duncan Jr
www.stephenduncanjr.com