We wanted to do something fun this hackathon! First, we thought of making a bot that imitates Trump’s speech patterns, generating sentences that sound like things Trump would say. Then we thought, "Let’s expand on that! Make bots that imitate all major presidents’ speech patterns, plus a few surprises!"
What it does
Two bots that are trained to imitate a famous person's speech patterns are randomly paired up and have a conversation about a random topic. Users then spectate for 60 seconds, and vote on who they think the bots are imitating.
How we built it
- Write web scraper to scrape online sources for corpuses of text (speeches, quotes)
- Generate Markov Chains for each person/bot
- Generate 100,000 sentences based on the Markov Chains
- Reverse index sentences based on topic
- Create web app
Challenges we ran into
How to efficiently reverse index 100,000 generated sentences
What we used
- Markov Chain
- Web-scraped corpus
- Speeches, quotes, scripts
- IBM Bluemix instance
- Flask, Django
What's next for Robot Flame Wars
- Launch on a webserver to the general public to use
- Feed user response back into the model--e.g. if users consistently misidentify a certain bot's real life analog, automatically adjust sentence generation as appropriate.