Comparison of Bayesian POP3 Spam Filters

kreide writes "Spam e-mail has become an ever increasing problem, and these days it is next to impossible to use e-mail without receiving it in large amounts. Although various techniques exits to combat the problem, spammers seemed to be winning the war - until a new, powerful weapon appeared on the scene: Bayesian filters, our last, best hope for spam-free inboxes. In this review I compare POP3 based bayesian spam filters." We did an Ask Slashdot on this a few weeks ago.

Comments

You just don't get it

[frovingslosh]

None of these spam filters will have any effect on spam at all if they are just installed on the systems of people who hate spam and would never buy from a spammer anyway. Hell, they might even have the opposite effect; I will never buy something if I get spam for it. But if I personally filter my spam and don't even see subject lines, I might end up buying the product without knowing they also marketed it by spam.

Spam is effective because it reaches millions of people who are not installing these filters on their systems. Until ISP's start applying these filters to all spam by default, then the spam filters will have no effect at all, exactly the same number of marks will be reached and respond no matter if the people who know better than to respond to spam go ahead and filter their e-mail or not!

Re:You just don't get it

[Plug]

Realistically, I don't give a damn how much spam _you_ get, I care that _I_ don't get any.

You cannot automatically filter spam. Bayesian filtering works because it works on your own personal items only, and you have a method of manually removing false positives. There is nothing worse than the possibility that an ISP will filter out a real email in their spam system. That simple fact makes server side spam filtering impossible for most situations. You can filter spam into /dev/null (unacceptable), you can filter into a spam box (How many POP users would that rule out, who only have one POP box?), or you can keep it bundled in email with a flag, and expect people to update their clients, in which case you have the exact scenario you have now - the client has to do something themselves.

Until Hotmail et al starts offering bayesian filtering with a separate 'spam' mailbox, consider server side filtering worthless.

I am smart and don't get any spam. A lot of people I see in my line of work, aren't. These people are going to get something like Outclass (an Outlook plugin for POPfile), and then they are going to see the problem go away, and they're not going to lose any email in the process.

I'd rather use SpamBayes, but the Outlook plugin has an annoying bug that renders autocompleting addresses in Outlook useless.

Re:Bayesian filters are useful, but...

[frovingslosh]

I still believe that we should have a hunting season for spammers, just like we do for ducks...

No, it should be longer, if not all year long.

Missing the point?

[aquishix]

As someone who recently acquired a B.S. in mathematics several days ago, I understand how these filters work. They are an excellent way to fight spam over the older methods.

However, I think that ultimately this sort of thing misses the point. Spam needs to be fought in the courts, not in the battlefield. I'm afraid that the success of these filters will cause spam NOT to become illegal, and thus lead to a world where we have a constant trickle of spam, albeit in small amounts.

I think we all agree that we want spam to be gone entirely, as is evidence by the first post being labeled as "troll" ;)

Filters do not stop spam...

[Tehrasha]

...they only prevent you from seeing it.

Your server and its harddrives still end up being a storage bin for it, and the spammers will continue to send as long as your machine allows it to be recieved. Always remember that spam differs from postal junk mail, in that the -receiver- pays for it. Unsolicited postage due mail.

Spam must be -blocked- and the ISPs that allow/encourage its continued spread must re-educated, or be put out of business. Only when spam becomes costly to send with it diminish.

The current proposed laws concerning the subject are currently focusing on content rather than consent. They dont mind if you get spammed with hundreds of ads, provided what is being advertised isnt fraudulent. They overlook the fact that the claim of you having 'opt in' for the spam is in itself the lie and fraud.

--Teh

Re:Bayesian filters are useful, but...

[dtfinch]

You know, computer crimes are considered terrorism under the USA PATRIOT Act. Until that silly law gets repealed, lets hunt down those terrorists for their, umm, denial of service attacks against innocent email users, bandwidth theft, failure to provide real opt-out links, sending email advertisements with fake return addresses, presenting obscene material to minors, etc...

Re:great

[devnulljapan]

Just remember though, we would never have television without commercials. Sometimes advertising is necessary.

NEVER?....Try the BBC?
No ads, quality programming, small fee.

Spam is not the same as commercial

[Eric Ass Raymond]

Please, go right on ahead and point out why spam is not the same as a commercial.

I'd be happy to.

I don't know about you but for me e-mail is an important part of my work - not something comparable to watching cable TV.

Spam clogs my mailbox and I have lost several important e-mails from clients when deleting the spam which, by the way, is often disguised as legitimate non-commercial mail and comes with forged headers. In addition to pushing fraudulent products, these facts make spam a completely different beast from the cable TV and its legitimate, controlled ads which eat up only my free time - not my emails or work efficiency.

"Bayesian"

[RDPIII]

I don't mean to troll, but I hope it's not too late to put an end to the unfortunate term "Bayesian spam filtering". This is perhaps the worst abuse of the adjective "Bayesian" I've seen, because nothing crucially depends on the application of Bayes' Theorem and/or on the use of Bayesian methods (informative priors, model selection, etc.). Why not simply call it "data driven spam classification" (as opposed to "rule based") or "empirical spam filtering"?

If the spam disaster had struck fifteen years ago, we'd all be talking about "neural spam filtering" (using artificial neural networks, ANNs) and basking in the warm fuzzy feeling imparted by the term "neural". But ANNs and Bayesian classifiers have the same interface: both are trained on labeled data and can be used to classify unlabeled data. The implementation details are not of primary importance, and if you think they are, I'd encourage you to look into large margin classifiers instead of Naive Bayes or ANNs.

You really just don't get it

[frovingslosh]

Realistically, I don't give a damn how much spam _you_ get, I care that _I_ don't get any.

But you still do get spam. Exactly as much of not more because you use Bayesian filtering. Spam still wastes your bandwidth to download that spam before it can be filtered. Spam still wastes any inbox size limits your ISP might impose. Spam cuts into any quota a forwarding service might now or in the future impose on your account, or it could take you to a higher charge level if you pay for a forwarding service. It costs your ISP money, costs that one way or another are eventually paid by you. Even the processing power for that Bayesian filtering costs you CPU cycles, while having no negative effect on the spammers whatsoever.

While you might not think you care how much spam I get, you might care if dozens, hundreds or thousands of other users at your work also get tons of spam, particularly when all of that spam significantly cuts into your bandwidth. And you will care when overload from spam on your mail server is so bad that it causes failures, effectively causing a D.O.S. situation.

And as long as geeks happly play with their little Bayesian filters, they stop seeing spam and so stop complaining to the providers that are letting spam get through. They stop doing other things that might make spammer's life difficult. Heck, I fully expect some spam haters with an additude like yours to say within earshot of a congressman or Senator something like "Oh, I never get any Spam. Spam can be filtered easily and nothing should be done about it". The spammers should love Bayesian filtering, it takes the presure off them while allowing them to reach exactly the same number of marks with a mailing.

Re:great

[impluvian]

I think there's a very simple distinction that can be made between spam and television advertising, and it has to do with the amount of control that your service provider exercises over the advertising content.

When you watch cable TV, you know that for an hour of content, you are going to see up to 12 minutes of advertising. The advertising is controlled by the cable company, and no-one can advertise on the channel without going through that 'filter'.

Spam, on the other hand, is not restricted. If I receive 100 e-mails a day, anywhere from 0 to 100 of them could be spam. None of those spams are sanctioned (or controlled) by my service-provider, and they were not part of the package I signed up for.

Why not stop the sellers?

[Anonymous Coward]

I know this is slightly off topic, but can someone answer me a reasonably simple question thats been bugging me for a while?

Why not instead of hunting down the spammers do we not hunt down the people who are selling and advertising their junk via the spammers?

The spammers purposly make themselves difficult to find, but it must be easier to track down a company that is collecting money and sending out products? Why not make the using of spammers services illegal and fine and punish those doing so?

I think Im correct in saying and please tell me if Im wrong, but here in the UK a similar situation is people "fly-posting". In these cases, if advertising posters are put somewhere illegal or unwanted, it is not the person who put the poster up that is fined, but the club, record label, whoever is beign advertised that takes the rap.

Just my 0.02p

Re:great

[Goldberg's Pants]

You probably ARE a scumbag spammer.

For people who have to pay for their online time (England for example), these scumbags are essentially stealing money from people. Filtering only works once you've downloaded the mail. You still have to download their worthless drivel. Sure, it may be pennies a week in costs for a user, but you tally that up over a year or two of dealing with these idiots, and you've got a sizeable chunk of change. Certainly enough for a nice pizza.

Let's not forget the TIME these shits waste as well. All this work invested in stopping spam. Who know's what cool stuff may have come from the minds who instead are working on ways of dealing with the email cancer.

As I said, these scumbags should be legal to hunt and kill.

Why filtering isn't the solution

[nuwayser]

An analysis of filtering methods against spam is kind of like a comparison of bullet-proof vests in that there's no incentive to stop someone from pointing a gun at you and firing it. In the past, spammers have been grossly affected by more sweeping changes, and I'm afraid filtering methods are only creating the mindset of, "Give up, use this software, it will do the deleting for you." It takes the attitude of, "just delete the stuff" and makes it automatic; sure it's convenient for a time, but in a year you're still going to get spam and your ISP will likely have fewer resources to deal with the complaints.

I'm saying, why not focus instead on technology which puts a bigger dent in spammers' ability to operate, like how to secure against proxy hijacking.

Re:A new *law* is required

[felis_panthera]

Out of that 2.2 million people, somewhere near 700,000 are in jail from possession, use or distribution of marijuana. A law that was originally used to control migrant mexican workers has bogged down the american legal system to the breaking point. Imagine, 700,000 new cells open for child molesters, rapists, spammers, and SCO executives.

Wouldn't it be grand?

PS: Sorry about the OT, but things like this need to be said whenever the opportunity presents itself.