Bradley Hayes from MIT has developed a Twitter-bot named @DeepDrumpf, which has the essence of Donald Trump’s locution and sends tweets. Bradley Hayes, postdoctoral associate at MIT has developed the bot which automatically tweets. Hayes, who is into robotics research, said that he has programmed recurrent neural network in order to study and emulate Trump’s speeches.
Hayes said that he was inspired to come up with something like this in part from John Oliver’s fantastic sketch. He shared that they were having a lunchtime conversation with his colleagues on various statistical modeling techniques that were vital for their research.
It was found that the same technique that has been used in DeepDumpf is quite useful in many robotics domains because it is a modelling technique that tries to learn the structure of sequential information. Natural language is the great example of sequential data.
Hayes has fed words into the algorithm taken from the transcripts of Trump’s debate performances and speeches after winning states in the primaries. The algorithm in return learns the underlying structure of the phrases and comes up with its own Tweets based on the structure.
Hayes said that Trump’s language tends to be more simplistic, which as per him was a modeling problem but would be the most manageable candidate to study. Hayes said that there is a lot more needs to be done in order to come up with a sample wall of text. So, then go through it and pick up the best 140-character post.
Hayes wants to make the process cleaner so that he can come up with more data. For this, he needs to interact with more candidates. Hayes is expecting that his development will start responding to the other primary candidates and could even deal with challenges, but that will take time.