According to the Annif-fusion experiments, using PAV a.k.a. isotonic regression could provide a significant improvement in analysis quality. We should add a new backend (called pav) that could be configured like this:
The parameter min-docs would specify the minimum number of documents about a subject for creating a PAV model, and the fallback parameter would define the strategy used when no PAV model exists. The values are either raw (use raw score) or zero (use zero score).
According to the Annif-fusion experiments, using PAV a.k.a. isotonic regression could provide a significant improvement in analysis quality. We should add a new backend (called
pav
) that could be configured like this:The parameter
min-docs
would specify the minimum number of documents about a subject for creating a PAV model, and thefallback
parameter would define the strategy used when no PAV model exists. The values are eitherraw
(use raw score) orzero
(use zero score).