Proven implementation of anti-spam in comments?

Question

Algorithms

It is necessary to develop an anti-spam for comments, machine learning is not suitable, since spam can change quickly and the machine will not have time to relearn. Has anyone tried to calculate text hashes, find the md5 intersection between comments, break words into hash sums and operate with them?
What rules were used?
As for links and phones, there are no questions, it's easier with them.

Answer 1

I tried to determine the roots of "parasitic" words.
If the percentage is greater than the threshold - spam.