Sort:  

No. I calculate it, by analizing the text itself.
It only works for English text thought...

Texts with non-latin letters (russian, chinese) have ~0-5 points. Non English texts with Latin letters usually get around 3-10 points. English posts written badly (terrible spelling) or having too many links and pictures and almost no text, usually get scores between 10 and 30.
I only reward posts with above 30 points.
The highest score I have seen is ~70 points.

How does FOSSbot work?

It's actually very similar to what you're describing, except that users can assign their own weights to the different factors and score accordingly.

You mean it also uses machine learning to analyse the language of the text? Damn!
I thought I was the first to come up with this idea...

I thought others would only analyze metadata, and not the text itself.

I don't recall the technical details, but it's on GitHub