I think the simple answer is "make it work like google"
. But that probably creates a lot of work.
I like the approach of higher score for files containing the greatest number of keywords. Then there could be a bonus for finding files that have the keywords in the proper sequence, e.g. first_second and second_third score higher than third_first.
This could be something like a bonus for each keyword matched, then a bonus for sequence matching (sort of like left-most match that is currently in FRR). I like the current score based setup, and the above fits within that framework.