Spam Problems in Help Desk > Any Solution? [Bayesian Filtering Doesn't Work]

Discussion in 'Using your helpdesk' started by Elite, Nov 25, 2011.

  1. Elite

    Elite Established Member

    Hello All,

    I'm using the latest stable build (4.30.649) of Kayako Fusion.

    Our help desk has horrible, horrible spam problems.

    For every legitimate support ticket we get, we probably get 10+ spam messages.

    Although we mark these messages as 'spam', trash them and ban the email address, it doesn't seem to be helping at all. (The spammers just keep changing their email address so a new ticket is created.)

    I know Kayako is supposed to have a Bayesian filter to help detect spam (read more), but has anyone actually gotten that to work?

    Right now, I'm inclined to believe it doesn't do anything at all. We've marked hundreds and hundreds of messages as spam, but it doesn't seem to be helping.

    When I go to Admin CP > Bayesian > Categories, my spam category has 1,018,951 words... so I assume it must be trained by now.

    Am I supposed to create another rule somewhere that checks the bayesian score and then flags the message as spam?

    I have read every bit of documentation Kayako has online and nothing seems to help.

    Does anyone have any ideas on how to get Bayesian to work or on any other spam fighting measures?

    Thanks in advance!
  2. Gary McGrath

    Gary McGrath Kayako Staff Staff Member

    Woah, thats a lot of words in your spam catagory :confused: I can show you how to get it filtering out spam, but when we do it, with that many words, its likely to treat every email you get as spam

    Basically if you goto admin cp, Bayesian, catagories, spam, increase the word weight. the higher you make it, the more and more "those words" are treated as spam. ( pretty sure its the higher you make it, I have not used it for awhile )

    Our spam catagory has 122 words in it, with a weight set at 500

    Also, in admin cp, Bayesian, settings, make sure its enabled.

    What you can do is goto, admin cp, Bayesian, diagnostics.

    goto the probably tab, copy and paste the contents of an email into the box, and press check, it will come back and tell you if it thinks this message is spam or not.

    Gary
  3. Elite

    Elite Established Member

    Thanks Gary!

    My Sys. Admin is actually going to clear out all the words we have in there so we can try to train it starting with a clean slate.

    Once he's done that, I'll double-check everything you suggested is implemented properly.

    Cheers!
  4. masterctrl1

    masterctrl1 Kayako Guru

    The default weight for Spam category is 1. Seems like that would be too low (.0001 probability?) to do anything based on the description of how this works.
    What about the No Spam setting?
  5. Igor Kryltsov

    Igor Kryltsov Established Member

    We do not have anything caught by Kayako Spam filter yet but I have a few obvious Spams every day - viagra etc. We have word count - 5,782 and weight 1. I guess we need to raise weight?
    Is this word count was taken from these emails I marked as Spam? How I can check these words? Do we need to reset system and start over?

    What happens when message identified as Spam incorrectly? Client will not get autoresponse until someone will move it to Inbox as not Spam?

    Thank you
  6. Gary McGrath

    Gary McGrath Kayako Staff Staff Member

    As far as I am aware, if the spam check triggers, it just ignores the email ( and logs it i think )

    I don't think you can see the list of words in the kayako admin cp, pretty sure its probably only viewable in the mysql tables.

    but yes, if you want it to trigger and block messages, you will need to increase the word weight on the spam catagory

    Gary
  7. Igor Kryltsov

    Igor Kryltsov Established Member

    I would expect it:
    - to learn when you say it is a spam when it was not caught (message and account created are automatically moved to trash)
    - to have preconfigured database of words and probabilities (how hard to take existing databases from SpamAssassin available online and include them into package?)
    - detected spam will still trigger email autoresponder just in case it was detected incorrectly (may be a different one saying - our spam filter stopped this message but we will examine it asap)
    - once you remove something from spam it learns again and recalculates probabilities

    So one does not need to take wild guesses about correct weights and worry about missing a valid support request.
  8. Igor Kryltsov

    Igor Kryltsov Established Member

    Does it automatically create a ticket and sends an autoresponder if you move it out of Spam?
  9. Igor Kryltsov

    Igor Kryltsov Established Member

    This Spam thing does not work in Kayako at all.

    I have 53,790 words in Spam category and category weight set to 700.
    As Gary said:"As far as I am aware, if the spam check triggers, it just ignores the email ( and logs it i think )"

    I do not want this because I want to see what was detected as Spam and be able to recover. But now ... Kayako never detects Spam. I hit Spam, than I delete a post, than I delete it from Trash, than I delete a registered user who sent spam. This is ANNOYING!!

    What an average support Joe would expect to have from a Product:

    - system detects Spam and puts it into Spam folder (autoresponder is not sent at this stage). I can not see Spam folder in Kayako at all. Does it exist? I mark a post as a spam and it stays in Inbox. Why?

    - average Joe checks Spam folder in a morning and can clear it with one button push. Not select check box and press button - this is two clicks! One click is a button "Delete all spam now!". All email clients do NOT put deleted Spam into Trash. Makes sense to follow them with this.

    OR

    - average Joe selects an email which was detected incorrectly as Spam and trains it as not spam with one button push - button 'Not Spam'. This SAME button trains Bayesian filter saying - "No, no. This one is not spam" AND moves a ticket to unnassigned queue AND sends autoresponder (processes it as it just arrived)

    In a mean time I entertain myself with "I hit Spam, than I delete post, than I delete it from Trash, than I delete registered user who sent spam."

    Igor
  10. masterctrl1

    masterctrl1 Kayako Guru

  11. Gary McGrath

    Gary McGrath Kayako Staff Staff Member

    Personally I think asking a helpdesk to employ a full anti spam solution is counter productive, we actually use our company wide anti spam service with our kayako helpdesk email addresses. Which gives the granular control mentioned above.

    A Bayesian filter is far from perfect, there are numerous ways to bypass it, there is even programs which will craft your email body to score low in a Bayesian test. That is why most anti spam services offer 7+ ways to detect spam to ensure they catch as much as possible.

    Gary
  12. Drew Keller

    Drew Keller Just one person in a world of millions.

    I agree with Gary, unless Kayako look to use the services of a cloud based solution, it will always be less effective than having a true spam solution.

    I would only rely on a helpdesk solution to do SPam filtering if it was intergrated with a real Spam solution and I actually dont see the point when there are some very cost effective cloud based spam filtering products out there.
  13. masterctrl1

    masterctrl1 Kayako Guru

    +100 -- option to integrate with other spam solutions through API?
  14. bs01

    bs01 Established Member

    Ideally spam filtering should be being done at the mailserver level. We get virtually no spam in Kayako because the mail server blocks 99,9% of it before it even gets picked up. Once a day we do of course check for false positives and up until now haven't had one.
  15. Igor Kryltsov

    Igor Kryltsov Established Member

    Guys,

    Kayako already has Bayesian Filtering and I just do not understand why it can not be used in a simple and handy way.
    When I test a spam email with this filter it detects with 99% probability as a Spam. So it works.

    But:
    - there is no spam folder in Kayako
    - as a result email can not be put in it
    - deleting spam is a many steps operation (could be one)

    I am not complaining about it not detecting or being incomplete on spam detection.

    If some people use a proper spam filtering solution - no problem. You check spam filter interface instead of doing it in Kayako and possibly it works better on detection part.
    I would prefer to use only one UI - Kayako as going to SpamAssasin, as example, will make it taking even more time and just remembering I have to do elsewhere a few times a day is too much in a busy small team operation.

    From my experience , I have got my inbox spam PROPERLY detected by Kayako's Bayesian Filtering, but mail still generates a ticket, registers a user and sits in inbox.

    All I am talking about is a few days of work - adding Spam folder and processing it in a way I described and NOT about improving spamability of current implementation.

    I think, you misunderstood my idea.
  16. Drew Keller

    Drew Keller Just one person in a world of millions.

    I disagree, I did not miss interpret your idea, I would just prefer Kayako work to deliver true helpdesk functionality rather than invest more time and effort into delivering something that they will never deliver to the degree of a true Spam solution.

    Having them deliver push alerts to mobile, desktop alerts, ticket reccurance management, integrated big screen, fixing the standard HTML parser, etc all have a much higher priority for myself as Spam is something that I don't feel will ever been fully managed by just using a Bayseian approach.
  17. Igor Kryltsov

    Igor Kryltsov Established Member

    Understand your point.

    Writing what I miss. This "Having them deliver push alerts to mobile, desktop alerts, ticket reccurance management, integrated big screen, fixing the standard HTML parser, etc all have a much higher priority for myself as Spam is something that I don't feel will ever been fully managed by just using a Bayseian approach." is not in my list of things I miss :)
  18. Igor Kryltsov

    Igor Kryltsov Established Member

    BTW is "fixing the standard HTML parser" is a fancy way of saying it will display HTML emails properly? I still can't beleive I have to ask clients to attach a table as a picture or XLS file as I can not see HTML emails.

Share This Page