-Re: Can I discover what part of a score is attributable to a subquery?
Paul Libbrecht 2012-04-14, 17:15
If I remember well, the big big problem is that there's all sorts of recalibration of the scores based on the query. Therefore having it in one go is really nice.
I am not sure the different similarity can be put together well here though...
Le 14 avr. 2012 à 18:58, Benson Margulies a écrit :
> On Sat, Apr 14, 2012 at 12:37 PM, Paul Libbrecht <[EMAIL PROTECTED]> wrote:
>> it was in the Lucene world in May 2010:
>> http://mail-archives.apache.org/mod_mbox/lucene-java-user/201005.mbox/%[EMAIL PROTECTED]%3E
>> Mark Harwood pointed me to a "FlagQuery" which was exactly what I needed.
>> His contribution sounds not to have been taken up, it worked for me in Lucene, 2.4.1.
>> We used this to create an auto-completion popup which selected the right language by flagging the right sub-query that was most matched.
> Paul, it seems to me that the criticism in the JIRA (do you really
> want this calculation for every single document that matches?) applies
> to me. In our stuff, we run a query, and we look at the top 200 items,
> rearranging their order based on a name similarity metric that is too
> expensive to run in bulk. If the overall query is 'just us', we can
> discard the Lucene scores and reorder based on our own. If our query
> is combined with other terms, then we need to subtract out the
> contribution our part of the initial query. However, sending in a
> second query with (I suppose) ids=id1,id2,... and just our query, to
> retrieve the scores, should be pretty speedy for a mere 200 items.
> Maybe I'm missing some even easier way, given a DocList and a query,
> to obtain scores for those docs for that query?
>> Le 14 avr. 2012 à 15:34, Benson Margulies a écrit :
>>> yes please
>>> On Apr 14, 2012, at 2:40 AM, Paul Libbrecht <[EMAIL PROTECTED]> wrote:
>>>> In mid 2009, I has such a question answered with a nifty score bitwise manipulation, and a little precision loss. For each result I could pick the language of a multilingual match.
>>>> If interested, I can dig.
>>>> Envoyé de mon téléphone Android avec K-9 Mail. Excusez la brièveté.
>>>> Benson Margulies <[EMAIL PROTECTED]> a écrit :
>>>> Given a query including a subquery, is there any way for me to learn
>>>> that subquery's contribution to the overall document score?
>>>> I can provide 'why on earth would anyone ...' if someone wants to know.