Google has attempted to make a comment moderation engine using ML, available at https://www.perspectiveapi.com/#/home
I don't know how well it works at the moment, it's typically been a very hard thing to do, since the amount of "data" in a comment tends to be small, and there is a lot of jargon and such in terms of context that can make different usage very complicated. My searches to remember that came up with some other alternatives in the same sort of space. As for "preferring to see them", that makes more sense when you receive very few of them. It's more complicated if your every post on say twitter generates dozens to hundreds of them. Most of them at that point are not "real", just offensive, and it seems unlikely that even humans would be able to tell the difference or deal with them at scale. One could imagine an automated system that collects the information in each and tries to group them by likely perps, using a scoring system over time to highlight the worse offenders. In a shared system, that could go even farther. In David Brin's _Earth_, one of the protagonists had an aside where they described a system that would scan their large email inbox and create a sentiment report highlighting with examples. I'm sure there are various systems like that probably tuned for brands and such on social media, and maybe even some for support desks. This use case is kind of a special case of that. Another alternative for such a system of detection is to have other people help go through the detritus for them, to relieve some of the dread that must come with constantly being bombarded with the cesspool. Anyways, I don't doubt that such crap isn't limited to social media, and that some makes its way to email... but I don't think there's been much work on it at that level. Brandon On Wed, Jul 15, 2020 at 2:15 PM Chris via mailop <[email protected]> wrote: > On 2020-07-15 16:09, Jaroslaw Rafa via mailop wrote: > > Dnia 15.07.2020 o godz. 12:59:39 Grant Taylor via mailop pisze: > >> On 7/12/20 11:28 AM, Paul Ebersman via mailop wrote: > >>> But in these tense times, lots of PoC/non-white/non-cis get hate > >>> emails, death threats, etc. > > >> Death threats seem like the warrant more effort to stop than spam. > > > If someone would send death threats to me, I'd rather would like to > receive > > them, so that I can know whether I'm actually in danger and optionally > > report to the police or whatever... > > > > Which is what I was about to say too. In fact, by being really good at > blocking harrassment/death threats, you become partly liable if > something does happen. Because you hid the threat from the victim, and > potentially increased their risk of harm from something they didn't know > about. > > I had to deal with some of these in a large corporate environment. Each > time our "reaction" to the threatening (or sometimes just crackpot) > emails was carefully considered between me (filter options) and > Corporate Security before deployment. Eg: > > - Do we reject/bounce, or silently ignore? Usually the latter. > - Do we forward captures to CorpSec, not the user? > - Law enforcement? Legal C&D? > > Etc. > > These also have to be carefully differentiated from some of the > wide-spread stuff like the "I have a contract on you, pay me, and I > won't kill you" extortion soan/scams. Those should just be nuked. > > I would generally not recommend filtering on "threatening content" > per-se, but on spam, however you usually do it - like DNSBLs or > spamassassin or whatever... Then make it a point of user education to > tell people what to do if they get threatening email that gets past the > general spam filtering. Which may be LE, may be your own corporate, may > be custom filtering, etc. This approach will generally tend to weed out > the normal broadcast bullshit, and leave behind the stuff the victim/you > really need to know about. > > _______________________________________________ > mailop mailing list > [email protected] > https://chilli.nosignal.org/cgi-bin/mailman/listinfo/mailop >
_______________________________________________ mailop mailing list [email protected] https://chilli.nosignal.org/cgi-bin/mailman/listinfo/mailop
