Not horrible, I guess, although I would have it warn and/or close the accounts automatically, and then have employees review appeals (ideally very quickly).
Otherwise I don't think that "initial surge" you mentioned would ever actually subside.
So banning the accounts is the end result? Aren't these throwaway accounts that can be recreated in a second? It seems like a reasonable concept to filter the Tweets from getting to me in the first place, I'm guessing a "spam box" is a little too complicated for Twitter though.
I took the report approach with an assumption (which like all assumptions could be extremely wrong) that pre-commit filters might not be something Twitter could implement for performance reasons at this time.
The report approach could eventually help with making the identification of spam easier, leading to more optimized approaches for filtering later.
Effective auto-moderator should check for multiple flags:
1) IP address (if it was banned in the last couple of months).
2) Registration Email address (was it banned in the past?).
3) Buzzwords (how frequently that word was used by banned accounts in the past)
4) Web sites (was it banned).
5) Age of the account (the younger - the more likely it's spam).
Based on all these red flags auto-moderator should calculate overall "spam rating", and above certain threshold accounts should be deleted automatically.
All key attributes of deleted account (IP, email, buzzwords) should be analyzed to train automoderator to recognize new spam).
I like that a lot, will help with the problem mentioned above of spammers just creating new accounts to some extent.
Another idea would be once I url is identified as spam, filtering it from ever being posted, and flag or suspend the account. Not sure how well Twitter can support pre-commit hooks for the tweets though.
Ironically, since this idea is on a Tumblr blog, I find Tumblr spam a lot more annoying. I help run a fairly popular Tumblr w/4K+ followers and occasionally I see that our posts are "liked" by users with names like "freebatteriesonline"; clicking on their name doesn't even take you t a Tumblr blog, just a dodgy battery sale site. The process of reporting spam Tumblr blogs is not intuitive -- the only instructions I found on the subject were on an eHow blog, for Pete's sake, and it's not clear to me whether I'm actually reporting them for spam or just blocking them so I can't see them.
I'm sure if twitter really wanted to fix the spam problem, they probably would/could have already. They seem to have the resources to tackle the problem, this poses the question: do they really want to? I'm sure on the business side of things, they measure various metrics, such as; account creation and tweets/minute/hour/day - they can then go to advertisers and tote such metrics. Perhaps tackling spam would actually hurt their bottom line? In the end, how bad does spam really affect you? I barely notice it, myself.
I live in a community governed by an HOA (even if I am on the board) I think pretty much ever facet of my life is a walled garden. Makes it easier to accept.
Is that open to anyone now? I thought the process was you developed against user streams then once you scaled beyond those you requested access to the stream api? And, you need to request access to the user streams api too right?
I decided not to do real time streaming for my current project so never followed up on it.
*edit: Actually looks like I'd want access to statuses/links which would solve the problem of filtering for links on my the scanning end.
There's no doubt twitter's useful. I use it myself daily. But I'd argue acceptance of the growing world of walled gardens is dangerous to innovation. I was sad reading your article because if this wasn't about twitter you could've written about how you SOLVED the spam problem, instead of begging a large company to smile upon your idea and maybe try it out.
Well, I supposed I could write a Twitter client that checked links and marked tweets as spam before presenting them to the user.
Sorry, it's my nature to try and solve problems working around whatever obstacles are in the way. So when your post made me think, well.... I could... In no way mocking or disagreeing with your opinion which I do comprehend and for a large part agree with.
On the same token I also think a truly open system, while not impossible, is a difficult proposition for financial reasons.
Also from a security stand point. If the system was open enough for me to solve the spam problem, then it would also be open enough to make the spam problem, or even virus/malware delivery much easier.
I think there may be a different approach: let spammers (aka bots) flourish and let users upvote the most helpful ones.
In short:
1. Create a second type of Twitter account for bots.
2. Give users the ability to opt out entirely of seeing bot replies.
3. Give users the ability to upvote helpful automated replies.
Offering spammers a regulated way of reaching a mass audience also gives Twitter the ability to be much more aggressive in shutting down accounts who pose as real users.
Isn't it a trivial task to automate twitter account creation? Shutting down christinanik2883 because she didn't register as a Botified Citizen doesn't do anything about jonnyi9032 or shellystack48799823.
I have an easier solution. Every time I get spam, I follow it through to the account. All of the tweets are identical: @<someuser> <filler text here> <identical url>
1. Scan all accounts
2. If you notice the above pattern, flag them for review.
3. In addition to #2, if X people flag them for spamming, suspend their account as well.
twitter is inherently a spammer's heaven. It's really easy to make 140 characters seem legitimate. An hour of writing spun content and a really easy to build signup bot and you can have full conversations, retweets, everything.
There's a reason twitter gets spammed so much, and it's because most of the time, normal people's 140 character messages look just as spammy as the actual spam.
28 comments
[ 4.3 ms ] story [ 71.6 ms ] threadOtherwise I don't think that "initial surge" you mentioned would ever actually subside.
That said, this seems like a reasonable idea.
The report approach could eventually help with making the identification of spam easier, leading to more optimized approaches for filtering later.
1) IP address (if it was banned in the last couple of months).
2) Registration Email address (was it banned in the past?).
3) Buzzwords (how frequently that word was used by banned accounts in the past)
4) Web sites (was it banned).
5) Age of the account (the younger - the more likely it's spam).
Based on all these red flags auto-moderator should calculate overall "spam rating", and above certain threshold accounts should be deleted automatically. All key attributes of deleted account (IP, email, buzzwords) should be analyzed to train automoderator to recognize new spam).
We implemented such system and it automatically deletes most of spam: http://postjobfree.blogspot.com/2009/06/postjobfree-automode...
Another idea would be once I url is identified as spam, filtering it from ever being posted, and flag or suspend the account. Not sure how well Twitter can support pre-commit hooks for the tweets though.
In the end spam == more traffic == more $$.
Another reminder that you're not using an open system, just someone's walled garden.
I decided not to do real time streaming for my current project so never followed up on it.
*edit: Actually looks like I'd want access to statuses/links which would solve the problem of filtering for links on my the scanning end.
Sorry, it's my nature to try and solve problems working around whatever obstacles are in the way. So when your post made me think, well.... I could... In no way mocking or disagreeing with your opinion which I do comprehend and for a large part agree with.
On the same token I also think a truly open system, while not impossible, is a difficult proposition for financial reasons.
Also from a security stand point. If the system was open enough for me to solve the spam problem, then it would also be open enough to make the spam problem, or even virus/malware delivery much easier.
In short: 1. Create a second type of Twitter account for bots. 2. Give users the ability to opt out entirely of seeing bot replies. 3. Give users the ability to upvote helpful automated replies.
I wrote a blog post with this suggestion a while ago: http://www.readwritehack.com/how-twitter-can-win-back-develo...
Hence bot-owners won't be that keen on officially registering as bots.
1. Scan all accounts 2. If you notice the above pattern, flag them for review. 3. In addition to #2, if X people flag them for spamming, suspend their account as well.
There's a reason twitter gets spammed so much, and it's because most of the time, normal people's 140 character messages look just as spammy as the actual spam.