224521 (1) [Avatar] Offline
#1
On page 104 There is a sentence dealing with the effect of synonym expansion on TF*IDF scoring that says: "...turning off norms all together so that all matches are scored equivalently no matter the term frequency or document frequency.

I think that turning off norms only affects length normalization (and index time boosting) but not the use of TF and IDF. See for example: http://lucene.apache.org/core/4_3_0/core/org/apache/lucene/search/similarities/TFIDFSimilarity.html

Do you mean that you want Lucene in constant-score mode?

Tom Burton-West
335818 (2) [Avatar] Offline
#2
Good find Tom, that definitely needs to be clearer. I meant that you might as well turn of the norms and term frequencies - but then of course you still have to contend with doc frequencies which can't just be turned off in elasticsearch. So yes, a constant-score is exactly the thing I should have said.

Thanks for the input!