Different relevance scoring in EL 2 and EL 5


(Jürg Mumprecht) #1

With almost identical settings / mappings / data in EL 2 and EL 5 I see different result ordering by "_score,desc". EL 2-Results are visually much better (documents containing all terms appear first; documents with more hits appear first).

I know that sounds not really precise. Is there any hint, what could be responsible for this new behaviour?

My Setup: Newspaper documents. I have separate indexes for languages and time-periodes; differences in mappings reflect only different language treatments.
Only difference in mappings from EL 2 to EL 5: _all is now enabled:false. which reduced the index-size dramatically.
Settings: Although it is now default I did not change similarity setting ("similarity": { "smdSimilarity": {"type": "BM25"}}.

Many thanks for any hint.
Best regards, Jürg


(Jürg Mumprecht) #2

Sorry folks; after deeper investigations I found a possible error in my mapping; it did not index parts of the text containing the tested query-words. Shame on me. I repair it and come back if the problem still persists.


(Alexander Reelsen) #3

Hey,

glad you found it. I moved the topic over the elasticsearch forum, as this is not an x-pack specific question. Please report back and help us track down possible issues, if you find them.

--Alex


(system) #4