The problem is that Twitter's filtered Streaming API has quite a small limit on the number of words that you can provide to match tweets. So, it would be imposs

The problem is that Twitter's filtered Streaming API has quite a small limit on the number of words that you can provide to match tweets. So, it would be impossible to scale. A better option might be retrieving tweets from the Firehose Streaming API, but one would need a massive backend to handle it. Additionally, access has to be requested to Twitter and it's very rarely granted. Alternatively, Datasift or Gnip can be better sources for the firehose, but I think they only provide a fraction (like 1/10) of the real amount of tweets that go through the actual firehose.