The problem with BM25 in a database, is that is can have unexpected outcomes for some common use cases.
Take multi-tenancy.
What if user 1 has many more documents than user 2, and uses "new orleans" a lot. But user 2 does not. User 2 does the search.
The db will first use FTS, and then filter. So user 1 will bias the results of user 2. Perhaps enough for user 2 to discover what words are in user 1 corpus.
Take multi-tenancy.
What if user 1 has many more documents than user 2, and uses "new orleans" a lot. But user 2 does not. User 2 does the search.
The db will first use FTS, and then filter. So user 1 will bias the results of user 2. Perhaps enough for user 2 to discover what words are in user 1 corpus.