yep, just use a customied similarity that doesn't include a length factor when computing the norm.
If you are currently using TFIDFSimilarity (or one of it's subclasses) then the computeNorm method delegates to a lengthNorm method, and you can override that to return "1" for fields with a certain name regardless of the length. If you are currently using something else -- like BM25Similarity perhaps -- you'll probably have to override the computeNorm method and write a slightly longer calculation based on whatever logic is in the computeNorm method you are currently using -- look for usages of FieldInvertState.getLength() and remove/replace that with a fixed value. : Date: Wed, 9 Mar 2016 13:23:16 -0500 : From: Matt Savona <matt.sav...@gmail.com> : Reply-To: java-user@lucene.apache.org : To: java-user@lucene.apache.org : Subject: disable field length normalization on specific fields? : : Hi all, : : I am trying to understand if the following is possible: : : I would like to have several fields in my index which are boosted at index : time. Because they are to be boosted at index time, their field type : requires omitNorms(false). : : However, I do not want field length normalization to affect the scoring of : these fields. For example, finding the term 'baseball' (1:5 words) should : score exactly the same as (1:100 words). : : There are other fields in my index which are not boosted, so : omitNorms(true) is acceptable on them. However, I do not want to broadly : disable length normalization on every single field (I have at least one : where I require it). Thus, I am not certain a custom Similarity class is : appropriate. : : Is it possible to simply disable length normalization on a a field-by-field : basis, while still allowing index-time boosting? : : Thank you in advance! : : - Matt : -Hoss http://www.lucidworks.com/ --------------------------------------------------------------------- To unsubscribe, e-mail: java-user-unsubscr...@lucene.apache.org For additional commands, e-mail: java-user-h...@lucene.apache.org
