Is the elasticsearch relevance score an ordinal or interval value?
In other words, is the difference between scores of 0.7 and 0.8 the same as the difference between scores of 0.5 and 0.6? Or is the 0.1 difference meaningless, not a measure of anything?
That's an interesting page but it doesn't answer my question, nor does saying it's "not a measure of anything".
its really not a measure of anything. Its just a value for ranking documents. Absolute value is meaningless and we take advantage of that in several ways: queryNorm, sum-of-logs vs multiplication, etc
I am not asking what it measures, I think we're talking at cross purposes.
I'd like to know if the scores are just ordinal, or if they are on an interval (or ratio) scale.
in general you can only assume: absolute values are meaningless. relative values are also meaningless.
If you care to understand more, then look at the actual formula for the similarity function you are using.