I have an idea, maybe not the best, but still an idea that should work.
First use a string replacer, your pattern will be your regex and replace the whole string using something that makes sense to you. Then simply use a groupby with a unique concatenate with count method.
Of course, this works only if you have 1 occurence of your regex in each row in the initial column.
Thank you for your suggestion. Finally I used 2 string manipulation nodes: one to match & replace the regex; and the second to count the replacements. This works also for multiple occurrences but I don't like because I consider it a dirty and dangerous workaround.
It's seems strange to me that in KNIME doesn't exist anything to do this in a clean way. Maybe we're missing something.