This isn't 50% of all tweets, it's 50% of all tweet impressions. That is, multiply each user's tweet by the number of users following them when they tweeted, and you'll find that 20K users (likely, people with more than a million followers who tweet with frequency) are half of all these impressions. However, the number of tweets you'd get by following the 20K would be far smaller than the firehose, probably a few hundredths of a percent at most.

It would be interesting, but it wouldn't be a representative 50% sample.

Why bother creating the accounts when you could just use RSS. Polling every minute should be sufficient.

http://twitter.com/statuses/user_timeline/259379883.rss

Winning.

Twitter won't allow that.

You could use the api

http://dev.twitter.com/pages/streaming_api

but then twitter would come after you if you tried to syndicate the content.

So then you have to deal with Gnip... and thats 10 cents/1000 tweets.

Is there a list somewhere? this sounds like an interesting idea.

20K

I believe you can follow 1000 people who don't follow you. So I think the Grandparent post was saying 20 accounts, each following a different 1000 people = 20K people followed, equals 50% firehose.

The twitter streaming api lets you track 5k users

http://dev.twitter.com/pages/streaming_api

so you'd only need 4 users and 4 clients running on 4 AWS micro instances.

But Twitter will shut you down once they find you syndicating the content.

2000. http://support.twitter.com/articles/66885-i-can-t-follow-peo...