lucene at mikemccandless
Feb 17, 2009, 3:33 AM
Post #8 of 8
(1710 views)
Permalink
|
It sounds like function queries would be a fit here -- they let you
customize how a documents score is computed, and eg you could blend
the "normal" Lucene relevance with your own contribution, per document.
Mike
sendtoprat [at] yahoo wrote:
>
> Ya. Earlier we were sorting only. But sorting didn't always result
> in optimal
> ranking as it completely ignores Lucene score. We do not want that.
> We want
> a combination of different features, and not necessarily sorting by
> one
> field values.
>
> Pratyush
>
>
> Michael McCandless-2 wrote:
>>
>>
>> Can you simply sort by that string field? (Which also uses
>> FieldCache
>> under the hood to get the values).
>>
>> Mike
>>
>> On Feb 12, 2009, at 3:46 PM, sendtoprat [at] yahoo wrote:
>>
>>>
>>> I guess I could use, String[] s =
>>> ExtendedFieldCache.EXT_DEFAULT.getStrings(reader, fieldname);
>>> It will consume memory but should not be that bad. For my index this
>>> field
>>> has sortable string value which I could use in my model. I only need
>>> to do
>>> this for one field. Comment ?
>>>
>>> Pratyush
>>>
>>> Michael McCandless-2 wrote:
>>>>
>>>>
>>>> [.You'd probably get more responses on java-user@ instead of
>>>> general@]
>>>>
>>>> Could you use per-field boost?
>>>>
>>>> Or, since it sounds like you have a separate field with the boost
>>>> you
>>>> want,
>>>> maybe you could do this all at search time using a function query?
>>>>
>>>> An advanced possibility is to make your own indexing chain, and
>>>> use a
>>>> different
>>>> NormsWriter, but that's an extremely big hammer to pull out for
>>>> this
>>>> nail.
>>>>
>>>> Mike
>>>>
>>>> sendtoprat [at] yahoo wrote:
>>>>
>>>>>
>>>>> Hi
>>>>> How can I set my own norm value during indexing time. From my
>>>>> quick
>>>>> reading
>>>>> of code, I believe that norm values are written by NormsWriter
>>>>> class
>>>>> which
>>>>> is called from final class DocumentsWriter. Norm values are set by
>>>>> calling
>>>>> org.apache.lucene.search.Similarity#lengthNorm(String fieldName,
>>>>> int
>>>>> numTokens). But I need to set the norm using the field value which
>>>>> is a
>>>>> float number. I will be using this norm for my retrieval model. I
>>>>> cannot use
>>>>> document.setBoost(), as i need to boost the fields differently.
>>>>>
>>>>> thanks
>>>>> Pratyush
>>>>> --
>>>>> View this message in context:
>>>>> http://www.nabble.com/Modify-Norm-tp21959177p21959177.html
>>>>> Sent from the Lucene - General mailing list archive at Nabble.com.
>>>>>
>>>>
>>>>
>>>>
>>>
>>> --
>>> View this message in context:
>>> http://www.nabble.com/Modify-Norm-tp21959177p21984666.html
>>> Sent from the Lucene - General mailing list archive at Nabble.com.
>>>
>>
>>
>>
>
> --
> View this message in context: http://www.nabble.com/Modify-Norm-tp21959177p21987878.html
> Sent from the Lucene - General mailing list archive at Nabble.com.
>
|
