> Applying filters hides the symptoms without treating the underlying illness, which is a society that is infested with sexism.
This doesn't follow from the symptoms being cited, which is that a small subset of people say incredibly offensive and inappropriate things online.
If a million people see a video, and 1% of the people respond, and 3% of the people responding are crazy, that's 300 crazy people responding. If each crazy person sends 15 emails before finding a new thing to obsess about, that's 4,500 threatening or otherwise crazy emails in response - which certainly seems like a flood to the person being responded to - but doesn't actually tell us anything about the population watching the video, at large, except that it has a similar percentage of crazy people to society at large.
Continuing the hypothetical: if you assume a normal person only sends 2 emails, then you only have 19000 normal emails to 4500 crazy ones, because crazy people are more talkative. At 1.2 emails for a normal person, It's 11000 to 4500, which makes crazy emails over a quarter of all messages received in response - a vastly disproportionate number considering they were 3% of respondents.
One thing that automated filters can provide is sentiment analysis and looking at related messages, to pull out the underlying distribution of people responding (and hence seeing it's just people being people), rather than just seeing the surface level "We have 25% crazy emails!"
I would have to see a much more detailed analysis of the emails, responses, and distribution of them to conclude that it's a widespread problem, rather than a small crazy contingent being amplified above their importance by just posting a lot.