Adaptive junk filtering

Today, I was wondering if the “Mark as Spam” button in Mailspring actually offers me any advantage over the “Move to Trash” button. What does “Mark as Spam” do in the background? It would be cool to not have to care about incoming junk/spam emails and make it learn the pattern of emails that the user deletes right away, so that they can get moved into spam/trash folders automatically.

Does this feature exist in another mail client or tool you use?

Yes, in Mozilla Thunderbird and I loved it for its accuracy. A link to an explanation can be found here: Thunderbird:Help Documentation:Dealing with Junk E-mail - MozillaWiki


(Originally posted by mhatvan on GitHub.)

Hey! Mailspring’s “Mark as Spam” option just moves the thread to your defined “spam” folder. I think that on Gmail and a few other providers the underlying mail server scans these emails and creates a model for blocking further spam, but older IMAP servers don’t do anything at all.

I don’t think we’ll invest much time in client-side spam / junk filtering in the future. Thunderbird’s client-side filtering was pretty good and we looked at pulling it or another open source project into Mailspring to do spam handling, but it’d be technically complicated because they require a lot of structured data and need to be deeply integrated into the app (to prevent things like notifications about spam emails.) It also turns out the majority of our users are connecting to mail services that perform spam filtering on their own (Fastmail, Gmail, etc.) at the mail server. Filtering server-side is much better because the volume of training data is so much higher and services can coordinate to track and identify trends.


(Originally posted by bengotow on GitHub.)

Thanks for this thorough explanation, totally understand :wink:


(Originally posted by mhatvan on GitHub.)

Years ago there was a Plugin for MS Outlook (yes, until 2010 Outlook was pretty usable for me) called “SpamBayes” (http://spambayes.sourceforge.net/)… it is written in Python and utilizes Bayesian Regression to classify mails as Spam or Not Spam. It wasn’t only extraordinary because of the classification algorithm, but also because of the perfect UI Integration which allowed to easily supervise the algorithm while learning until it was doing well (after manually selecting 10 to 20 Mails manually I hardlly had any misclassification which I could then use to retrain the algorithm)

Long story short…it was working like a charm.

I would love to see a plugin like that for mailspring again which really would be a major feature I missed the last 10 years or so in any mail client I tried (Outlook since 2013, Thunderbird, Evolution, Gmail, …). If you already know a solution like that please share.


(Originally posted by pinkerltm on GitHub.)