4 ms·
One thing I miss in Thunderbird is automatic labelling of e-mails. Way back I used to use Opera's e-mail client, which I think was called M2. It had bayesian le
by zelphirkalt 3y ago
One thing I miss in Thunderbird is automatic labelling of e-mails. Way back I used to use Opera's e-mail client, which I think was called M2. It had bayesian learning for labelling. You would label e-mail and then it would learn from that and label new e-mails itself quite reliably after some training. Then you could build filters for.
In Tb one can tag e-mails, but I have yet so see a good way to have it learn this and automatically label e-mails. Assigning tags is still somewhat cumbersome via the context menu or the tags bar, which unfortunately disappears when selecting multiple e-mails.
- Koffiepoeder 3y agoYou can create manual automation rules through the 'filters' functionality though [0]. It's not automatic/bayesian based, but after some tinkering I manage to use it quite effectively. [0]: https://support.mozilla.org/en-US/kb/organize-your-messages-using-filters https://support.mozilla.org/en-US/kb/organize-your-messages-...
- zelphirkalt 3y agoI know those and sometimes use filters, but am too lazy to manually create filters for all my tags. Also one issue is, that the filters can only either be "or" or "and" filters. One cannot make some conditions "or" and the result of those "and" with another condition. Except if you utilize several layers of folders, having filters working on different layers. At that point it becomes a whole algorithm.
- yobbo 3y ago"Bayesian" in this context most likely means naive bayes; which assumes occurrences of all words are independent of each other. The "score" of each label ends up being something like the product of all relative word frequencies, multiplied by the probability of the label itself: p(L) * p(W|L) / p(W). But this would be super-effective if header fields like sender addresses and subject lines were somehow used as separate features, and you wanted to label based on that.
- zelphirkalt 3y agoYeah, or even be able to select what constitutes a feature for naive bayes and what is not considered. For each label separately configurable!
- pmlnr 3y agoOpera's mail client was wonderful, and there's nothing close to it right now. I had high hopes for Geary at one point, but it way as unstable as it gets the last time I tried.
- Nux 3y agoHave you tried Vivaldi's email client?