lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Francisco Sanmartin <francis...@olx.com>
Subject Re: More Like This boost
Date Tue, 22 Apr 2008 17:24:36 GMT
Yep, it would be nice for  MLT to have this feature, that's why I am 
trying to do it from the querys before sending the query to Solr. These 
are the steps I'm following:

1. execute a mlt.like() with the text document_example.getTitle() 
against the field "Title" of all the other documents. This returns a 
query containing the most relevant words in the example_document and in 
the rest of documents in the Title. We will call this query 
"QueryTitle". For example QueryTitle = (words^0.4 in^0.3 the^0.56 
title^0.65)
2. execute a mlt.like() with the text document_example.getDescription() 
against the field "Description" of all the other documents. This returns 
a query containin the most relevant words in the example_document and in 
the rest of documents in the Description. We will call this query 
"QueryDescription". For example QueryDescription = (other^0.66 words^0.7 
in^0.33 the^0.49 description^0.43)

Up to here, everything is possible with the options that offers MLT.

Now, with the info MLT gave me (QueryTitle and QueryDescription), i want 
to look in Solr for the documents (and more filters) to retrieve the 
best matches. But I want QueryTitle to be more important that 
QueryDescription, for example 70% and 30% respectively. This means that 
we should do QueryTitle^0.70 and QueryDescription^0.30. This means 
having a query for Solr like this:
(words^0.4 in^0.3 the^0.56 title^0.65)^0.70 (other^0.66 words^0.7 
in^0.33 the^0.49 description^0.43)^0.30

The question is...is Solr able to "understand" a query boosted who has 
its terms boosted already? (Remember that MLT returns the "interesting 
terms" boosted). This does make sense? Will the words obtained from a 
mlt.like() on the title be 70% relevant while the words obtained from a 
mlt.like() on the description will be only 30% relevant?

Of course it would be a nice feature to be able to boost these things 
natively and do only one call to MLT...Don't hesitate to contact me if 
you need any help on developing this feature.

Thanks!

Pako

Erik Hatcher wrote:
> No, the MLT feature does not have that kind of field-specific boosting 
> capability.  It sounds like it could be a useful enhancement though.  
> Of course you do get boosts for "interesting terms" already, but maybe 
> having an additional field-specific boost would be a nice touch too.
>
>     Erik
>
> On Apr 22, 2008, at 9:13 AM, Francisco Sanmartin wrote:
>> I know that only one query of that type does not change anything. But 
>> when it's two or more with different boosts, i hope it does. Here is 
>> the situation:
>> My docs have "Title" and "Description". What I want to do is to give 
>> more relevancy to the morelikethis on the title than on the 
>> description. So the query would be like this:
>>
>> query = (words^0.4 in^0.3 the^0.56 title^0.65)^0.70 (words^0.7 
>> in^0.33 the^0.49 description^0.43)^0.30
>>
>> This way, the words in the title are more relevant than the words in 
>> the description, right?
>>
>> Thanks!
>>
>> Pako
>>
>>
>> Erik Hatcher wrote:
>>>
>>> On Apr 21, 2008, at 5:02 PM, Francisco Sanmartin wrote:
>>>> Is it possible to boost the query that MoreLikeThis returns before 
>>>> sending it to Solr? I mean, technically is possible, because you 
>>>> can add a factor to the whole query but...does it make sense? 
>>>> (Remember that MoreLikeThis can already boosts each term inside the 
>>>> query).
>>>>
>>>> For example, this could be a result of MoreLikeThis (with native 
>>>> boosting enabled)
>>>>
>>>> queryResultMLT = (this^0.4 is^0.5 a^0.6 query^0.33 of^0.29 
>>>> morelikethis^0.67)
>>>>
>>>> what I want to do is
>>>>
>>>> queryResulltMLT = (this^0.4 is^0.5 a^0.6 query^0.33 of^0.29 
>>>> morelikethis^0.67)^0.60      <---(notice the boost of 0.60 for the 
>>>> whole query)
>>>
>>> That last boost wouldn't change the doc ordering at all, so it'd be 
>>> kinda useless.
>>>
>>> What are you trying to accomplish?
>>>
>>>     Erik
>>>
>>>
>
>


Mime
View raw message