Is it possible to programmatically 'clean' emails?

Question

Does anyone have any suggestions as to how I can clean the body of incoming emails? I want to strip out disclaimers, images and maybe any previous email text that may be als

Answer 1

SigParser has an assembly you can use in .NET. It gives you the body back in both HTML and text forms with the rest of the stuff stripped out. If you give it an HTML email it will convert the email to text if you need that.

var parser = new SigParser.EmailParsing.EmailParser();
var result = await parser.GetCleanedBodyAsync(new SigParser.EmailParsing.Models.CleanedBodyInput {
    FromEmailAddress = "john.smith@example.com",
    FromName = "John Smith",
    TextBody = @"Hi Mark,
This is my message.

Thanks
John Smith
888-333-4434"
            });

// This would print "Hi Mark,\r\nThis is my message."
Console.WriteLine(result.CleanedBodyPlain); 

Answer 2

Given your question "Is it possible to programmatically ‘clean’ emails?", I'd answer "No, not reliably".

The danger you face isn't really a technological one, but a sociological one.

It's easy enough to spot, and filter out, some aspects of the messages - like images. Filtering out signatures and disclaimers is, likewise, possible to achieve (though more of a challenge).

The real problem is the cost of getting it wrong.

What happens if your filter happens to remove a critical piece of the message? Can you trace it back to find the missing piece, or is your filtering desctructive? Worse, would you even notice that the piece was missing?

There's a classic comedy sketch I saw years ago that illustrates the point. Two guys working together on a car. One is underneath doing the work, the other sitting nearby reading instructions from a service manual - it's clear that neither guy knows what he's doing, but they're doing their best.

Manual guy, reading aloud: "Undo the bold in the centre of the oil pan ..." [turns page]

Tool guy: "Ok, it's out."

Manual guy: "... under no circumstances."

Answer 3

A few obvious things to look at:

1. if the mail is anything but pure plain text, the message will be multi-part mime. Any part whose type is "image/*" (image/jpeg, etc), can probably be dropped. In all likelyhood any part whose type is not "text/*" can go.
2. A HTML message will probably have a part of type "multipart/alternative" (I think), and will have 2 parts, one "text/plain" and one "text/html". The two parts should be just about equivalent, so you can drop the HTML part. If the only part present is the HTML bit, you may have to do a HTML to plain text conversion.
3. The usual format for quoted text is to precede the text by a ">" character. You should be able to drop these lines, unless the line starts ">From", in which case the ">" has been inserted to prevent the mail reader from thinking that the "From " is the start of a new mail.
4. The signature should start with "-- \r\n", though there is a very good chance that the trailing space will be missing.

Answer 4

If you creating your own application i'd look into Regex, to find text and replace it. To make the application a little nice, i'd create a class Called Email and in that class i have a property called RAW and a property called Stripped.

Just some hints, you'll gather the rest when you look into regex!

Answer 5

Version 3 of OSBF-Lua has a mail-parsing library that will handle the MIME and split a message into its MIME parts and so on. I currently have a mess of Lua scripts that do stuff like ignore most non-text attachments, prefer plain text to HTML, and so on. (I also wrap long lines to 80 characters while trying to preserve quoting.)

As far as removing previously quoted mail, the suggestions above are all good (you must subscribe to some ill-mannered mailing lists).

Removing disclaimers reliably is probably going to be hard. My first cut would be simply to maintain a library of disclaimers that would be stripped off the end of each mail message; I would write a script to make it easy for me to add to the library. For something more sophisticated I would try some kind of machine learning.

I've been working on spam filtering since Feb 2007 and I've learned that anything to do with email is a mess. A good rule of thumb is that whatever you want to do is a lot harder than you think it is :-(

Answer 6

In email, there is couple of agreed markings that mean something you wish to strip. You can look for these lines using regular expressions. I doubt you can't really well "sanitize" your emails, but some things you can look for:

1. Line starting with "> " (greater than then whitespace) marks a quote
2. Line with "-- " (two hyphens then whitespace then linefeed) marks the beginning of a signature, see Signature block on Wikipedia
3. Multipart messages, boundaries start with --, beyond that you need to do some searching to separate the message body parts from unwanted parts (like base64 images)

As for an actual C# implementation, I leave that for you or other SOers.