On Mon, Sep 19, 2022 at 2:39 PM Slavko via mailop <mailop@mailop.org> wrote:
> Dňa 19. septembra 2022 20:45:27 UTC používateľ Brandon Long < > bl...@google.com> napísal: > > >The simple answer is you add that signal to the list of other signals in > >your machine learning > >model and let the model training figure out how useful it is as a signal > >and what to combine > >it with. Depending on the type of ML, you may or may not be able to see > in > >the model the > >utility of the specific signal. > > I did some experiments with rspamd's neural module recently, but the > results had too high false positives (up to 25 %). I cannot decide, if its > model fails or here was low messages count in learn stage or i configure > it wrong, or whatever else. But with this result, it is not reliably usable > for me, and i cannot build anything based on its results. > > I am not aware of any other thool, which one can "simple use" in > mail processing. And to learn it from scratch and to build my own > model/tool i have not enough time nor knowledge nor experiences in > that. > > Anyway, i do not believe the machine learning systems mostly, when i > consider that it will have bugs in code, bugs in models (as any SW > has) and mistakes in learning, that is not something what i want > to rely on it, despite that it is big bussword nowadays. > Yes, this assumes that you're already using a machine learning system and have enough ham & spam to feed it, and ongoing signals and re-training and all of that. It is certainly not trivial to get working. Using machine learning for personal mail or very few users, not sure if that's likely to be worth the investment unless you want to learn a lot about it. You are likely to have a good history of "good" mail in an archive, at least, which means a model to learn what to whitelist may work well. Brandon
_______________________________________________ mailop mailing list mailop@mailop.org https://list.mailop.org/listinfo/mailop
