I know that in the end it, can't be done.
But, what are the options to:
a) limit the options for persons to create multiple accounts,
b) increase the chance of detecting multiple accounts / person
for a blog-like web service?
(people can sign up for their own blog)
Update:
I think the 'limit the options' has been answered nicely. (there is no reliable method, but we can raise the bar)
However, I would still like to know what other options there are to detect multiple accounts?
One common option is to verify the persons identity through their e-mail. Actually make them respond to an e-mail sent to their account. Some sites take this a step further and don't allow addresses from domains such as yahoo, g-mail, hotmail, etc ...
I think the best method would be to remove the incentives for creating multiple accounts.
Do you limit the users in any way? Can those limits be overcome (easily) by creating multiple accounts? If so, then maybe you should think about removing those limits.
You can't and you shouldn't. You are not dealing with the real world guys, but with accounts, so treat them as abstract entities which have the equal rights to live.
Some options I can imagine on the fly:
-- Only one account for email address. But I can create more then one email... or use Mailinator.
-- Long and tedious verification procedure. But that will discourage the users from registration
-- bind the IP to the account and block(temporarily?) that IP from creation of another account. But two different users with the same gateway will be blocked...
-- Use the cookies. But the user can delete them.
I think an alternative direction to take with this is to let the "big boys" do it.
http://oauth.net/
Offload the authentication of your site to a well-known 3rd party like Google or Facebook. It won't prevent duplicate accounts, but it's nice to think that the latest in spam prevention and whatnot is automatically implemented for you.
You could send users a SMS message to verify before creating the account. Since people can't get cell phone numbers as easily as they can get email addresses, this might work. Some people might be able to get two or three accounts, but not an unlimited number. There are a number of services that let you send SMS messages programmaticly, including Gizmo SMS, Text4Free and TxtDrop.
Of course, this requires users to have cell phones, and be willing to provide you with the number.
I'm assuming you're talking about a free service? I can't think of any ways that don't either have serious drawbacks or would be trivial to defeat. Things like setting a cookie, requiring a unique e-mail address are easy to defeat.
Requiring a unique IP address is not foolproof but might work to some degree, up to the point that you have lots of users and get complaints from people behind proxies.
The best ways are to charge money or require people provide some kind of personal information, like real name/phone/address that you verify, or a CC number, but that's invasive (then again maybe you only want serious users who are willing to provide this sort of info).
I guess I would turn the question around and ask "Why don't you want to let people have multiple accounts?"
There may be some other ways of mitigating whatever your underlying reason is, i.e. if you're worried about lots of orphaned blogs you could scan for a period of inactivity and disable them or at least schedule them to be looked at by a human. If you're worried about spam blogs you could periodically scan all blog content for spammy stuff. If you're worried about bots and are using some generic software like WordPress, change the names of the form variables and otherwise protect your forms from bots.
Definitely think of other ways of dealing with the problem, because you are not going to be able to block people from registering multiple accounts if it's a typical free service like Blogger.
As for detecting multiple accounts by one person, the first thing you need to do is have a log file store complete data on every user login (username, timestamp, IP, user-agent etc.), that you can then analyze later. I'll list a few things to look out for, but just by poring over the log file you will likely discover other patterns. Some ideas of things to look for are:
If you're worried about spam blogs, you could try doing some analysis of blog content, i.e. extract all the
<a href>
s and look for correlations between blogs. You could run the blog content itself though something like SpamAssassin or otherwise filter for spammy words like "viagra" and "rolex."