pavel.goncharik at gmail
Mar 27, 2012, 2:58 PM
Post #3 of 4
(171 views)
Permalink
|
Dear Lucene users and developers,
sorry for getting back to this old subject, but we are in the position
of re-evaluating our current implementation, which uses re-compiled
version of Lucene 3 with boolean scorers multiplying sub-scores. I was
hoping that "flexible ranking" in Lucene 4 will provide a mechanism
for that, but after a brief look it seems that "adition" is still
locked in corresponding implementations of boolean scorers. I also can
not see how CustomScoreQuery, suggested below, might help, so here we
go again.
To illustrate the problem with a simplified example, think of a
boolean query, comprised of 3 sub-clauses (A and B and C):
then if for document #1 the score for query A = 3, for B = 3, for C =
3, then result score 3 + 3 + 3 = 9;
and for another document #2 where score for A = 8, for B = 1 and for C
= 1, result score 8 + 1 + 1 = 10.
The thing is that in my case doc #1 is much more important, because
its sub-clauses are "represented" much more evenly than in the doc #2,
where B and C are represented poorly and it bubbles up only because of
over-blown (for whatever reason) score of A. What's needed is some
mechanism to represent poorness of B and C stronger - e.g. simple
multiplication instead of addition, so that the score for doc #1 =
3*3*3 = 27 and for doc #2 = 8*1*1 = 8, and doc #1 wins.
1) do I miss something in Lucene 3 or 4?
2) wouldn't it be much more flexible, if there was an easy-to-override
method/parameter in e.g. Similarity or BooleanQuery, to be invoked by
boolean scorers for calculating compound score of boolean sub-clauses?
On Tue, Jul 26, 2011 at 1:00 PM, Ian Lea <ian.lea [at] gmail> wrote:
> Have you tried CustomScoreQuery/CustomScoreProvider? Complicated but powerful.
>
>
> --
> Ian.
>
>
> On Mon, Jul 25, 2011 at 9:29 AM, Pavel Goncharik
> <pavel.goncharik [at] gmail> wrote:
>> Hi,
>>
>> as far as I can see, boolean scorers always sum up scores of their
>> sub-scorers. It works, but in case of my application it's required to
>> multiply sub-scores.
>> Is there a simple/efficient way to do this (apart from modifying
>> lucene's source code)?
>> It seems to me that standard tricks (e.g. CustomScoreQuery or custom
>> Similarity) would not help me in this case.
>>
>> Thanks in advance for helping!
>>
>> - Pavel
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: java-user-unsubscribe [at] lucene
>> For additional commands, e-mail: java-user-help [at] lucene
>>
>>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: java-user-unsubscribe [at] lucene
> For additional commands, e-mail: java-user-help [at] lucene
>
---------------------------------------------------------------------
To unsubscribe, e-mail: java-user-unsubscribe [at] lucene
For additional commands, e-mail: java-user-help [at] lucene
|
