What is sarcbot?
It's a twitter bot that posts snarky remarks about stocks that are on every trader's mind. It pays particular attention to those stocks that have incredibly high one-sided sentiment, like 90% bullish, or something.
How does it work?
It loves StockTwits and the Yahoo message board. When it finds a stock of interest, it confirms that it has one-sided sentiment. Then it goes troll fishing on the Yahoo message board for instigators. Recent comments are then fed through an NLP classifier to determine if it belongs on the feed. Bot or Not? One bot-worthy comment is selected at random for each stock on the list. After a random amount of time, @sarcbot spits it out.
How'd you get a training set?
Let me start by saying, if you spend a significant amount of time on the Yahoo message boards, ... you're hurting your brain. I built my dataset. I read through 1000s of messages and decided which messages I would like to see on the stream, and which ones I would not like to see. I am dumber for it.
1) The bot is currently in a hacked together form.
2) Its accuracy is around 70%.
3) It uses 2 different machines because one of them is too old to install numpy.
4) There's a hidden web service for pushing features and classification results between one-another.
5) This is my first bot with a machine learning algorithm behind it.