To be clear, I wouldn't filter people just because they have different views than me (the goal is to automate the detection, to avoid the effort of reading all the comments -- I should mostly not be in the loop). But I have come across accounts that openly admit to being sock-puppets (eg https://news.ycombinator.com/item?id=47242156). These sorts of accounts I would highlight.

Likewise for guideline-abusers. I don't really know what heuristic you would use to detect rules abuse, but I imagine there are at least some clear violations that could be detected.

Finally, I think I'd make one account for sock-puppets, another for guidelines-abusers, etc, so people can 'subscribe' to whatever degree of 'highlighting' that they want.