What does the "Offensive Language Model False Positive Score" represent?

[atdt1]

I am seeing various messages being flagged with a "Offensive Language Model False Positive Score" with varying ratings from 0 to 50. What do these scores mean? I have reviewed some messages with high scores and they look fine. 

Since the scoring is for "false positive", are lower scores more concerning?

[#2]

Hello,

i have the same problem understanding how such model works (especially the False Positive)

Regards.