TrumpOfYore: Using AI and math to find similar tweets from the president’s past

As noted in the writeup, sample size is crucial for text data, particularly for text which is idiosyncratic (and Trump's tweets are very idiosyncratic). And also as noted, the modeling is from 2013, and internet lingo has evolved significantly since then. (it's also a reason I hate pretrained word vectors like GloVe from around then).

Fortunately, Reddit has millions and millions of contemporary submission titles available in a publicly available and easily downloadable format. :)

Tldr: it's built using Dhingra et al.'s Tweet2Vec system - https://github.com/bdhingra/tweet2vec to convert tweets into word vectors and then looks at the cosine similarity to find similar tweets of yore