Static Quality Scores and Boost Functions


(Stephen Day-2) #1

An integral part of quality search results comes from ranking through
static quality scores, described here in Manning's IR book:

http://nlp.stanford.edu/IR-book/html/htmledition/static-quality-scores-and-ordering-1.html

In SOLR, one is able to affect ranking through boost functions,
applied to the final query scoring, which can work to be a kind of
"online" quality score. I have scoured the elasticsearch
documentation, but have not been able to find similar functionality.
Does elasticsearch not have this capability? Is this a planned
addition?

Thank you,

steve


(Shay Banon) #2

Boost functions are certainly on the roadmap. You can also plug in your own
Similarity (Lucene) implementation to further control the scoring. Its not
documented (yet), but its there. So if you want that, I can point you on how
to do it.

-shay.baon

On Mon, Mar 22, 2010 at 3:50 AM, Stephen J Day stevvooe@gmail.com wrote:

An integral part of quality search results comes from ranking through
static quality scores, described here in Manning's IR book:

http://nlp.stanford.edu/IR-book/html/htmledition/static-quality-scores-and-ordering-1.html

In SOLR, one is able to affect ranking through boost functions,
applied to the final query scoring, which can work to be a kind of
"online" quality score. I have scoured the elasticsearch
documentation, but have not been able to find similar functionality.
Does elasticsearch not have this capability? Is this a planned
addition?

Thank you,

steve


(Stephen Day-2) #3

Thanks for the response on this, Shay.

Glad to hear that Boost Functions are on the road map and am quite
interested to see what you come up with. Do you plan on following suite with
SOLR, or integrating some into the Query DSL? Do you have a grammar for the
query DSL? It might be interesting to define the scoring in the Query DSL,
especially if you formalize the grammar.

I have worked a little bit with Lucene Similarity classes, but if you have
the time, documenting the process would be appreciated.

steve

On Mon, Mar 22, 2010 at 2:22 AM, Shay Banon shay.banon@elasticsearch.comwrote:

Boost functions are certainly on the roadmap. You can also plug in your own
Similarity (Lucene) implementation to further control the scoring. Its not
documented (yet), but its there. So if you want that, I can point you on how
to do it.

-shay.baon

On Mon, Mar 22, 2010 at 3:50 AM, Stephen J Day stevvooe@gmail.com wrote:

An integral part of quality search results comes from ranking through
static quality scores, described here in Manning's IR book:

http://nlp.stanford.edu/IR-book/html/htmledition/static-quality-scores-and-ordering-1.html

In SOLR, one is able to affect ranking through boost functions,
applied to the final query scoring, which can work to be a kind of
"online" quality score. I have scoured the elasticsearch
documentation, but have not been able to find similar functionality.
Does elasticsearch not have this capability? Is this a planned
addition?

Thank you,

steve


(Shay Banon) #4

Hi,

I am actually not familiar with Solr boost functions, I wrote my own
version of the current function query in Lucene about 5 years ago :). I
actually have some really nice ideas about how to implement it properly. It
will probably include a DSL, but I want to make it even simpler. Still
thinking about it thought as I need to create some support for it in
elasticsearch code...

-shay.banon

On Wed, Mar 24, 2010 at 12:16 AM, Stephen Day stevvooe@gmail.com wrote:

Thanks for the response on this, Shay.

Glad to hear that Boost Functions are on the road map and am quite
interested to see what you come up with. Do you plan on following suite with
SOLR, or integrating some into the Query DSL? Do you have a grammar for the
query DSL? It might be interesting to define the scoring in the Query DSL,
especially if you formalize the grammar.

I have worked a little bit with Lucene Similarity classes, but if you have
the time, documenting the process would be appreciated.

steve

On Mon, Mar 22, 2010 at 2:22 AM, Shay Banon shay.banon@elasticsearch.comwrote:

Boost functions are certainly on the roadmap. You can also plug in your
own Similarity (Lucene) implementation to further control the scoring. Its
not documented (yet), but its there. So if you want that, I can point you on
how to do it.

-shay.baon

On Mon, Mar 22, 2010 at 3:50 AM, Stephen J Day stevvooe@gmail.comwrote:

An integral part of quality search results comes from ranking through
static quality scores, described here in Manning's IR book:

http://nlp.stanford.edu/IR-book/html/htmledition/static-quality-scores-and-ordering-1.html

In SOLR, one is able to affect ranking through boost functions,
applied to the final query scoring, which can work to be a kind of
"online" quality score. I have scoured the elasticsearch
documentation, but have not been able to find similar functionality.
Does elasticsearch not have this capability? Is this a planned
addition?

Thank you,

steve


(system) #5