One of the fields in our search is an array of person names. We see that when searching for a the name of a single person, documents with several people attached score lower. From what I can tell this is because the document length dl in the BM25 algorithm is accumulated over the whole list of people.
Is there a way not to score documents with several people lower? For now my only idea is to set b=0 for the person field and so discount document length altogether, although I expect this not to be ideal – we’d still like to take into account document length within each person name.
Note that our full search is a bool search with several multimatch queries etc, but the behaviour described above is retained when paring it down to searching the person field only.