Donald Trump or Hillary Clinton? Nope, this is not what you think... I have no intention to write about politics. This post is about machine learning.

I once built a machine learning model that would tell you whether a given tweet belongs to President Donald Trump or Hillary Clinton. The idea was to practice and apply the theory about Recurrent Neural Networks (RNN). The intuition behind the RNN, in my opinion, is that it introduces the concept of "context". When you read this text, for example, you don't think about every single word separately. You rather think cumulatively. You consider the words that were before and build up the message that I am trying to deliver as you go towards the end of this text. In other words, the sequence of words itself bears fundamental information. Conventional neural nets don't allow you to capture this. RNN was introduced to solve this.

There have been several upgrades of RNN over the years. In this project, I built three types of RNN, trained them on over 4000 tweets and compared the results. The models that I built were typical ones: vanilla, LSTM, and GRU. GRU was the best by predicting the owners of the tweets with 95% accuracy. I left the link to my blog post in the bio section where I explain the project in minute details.

I gave the following tweet to my model: "The question in this election: Who can put the plans into action that will make your life better?" My model predicts with 93.6% probability that this tweet belongs to _______. Wait... let's see who is better: you or the model.