Paris is to France as Berlin is to ?

Car is to road as train is to ?

King is to queen as husband is to ?

These all seem like really easy analogies, and they are for humans, at least. For machines, not so much. That's why I tried creating a little program capable of taking advantage of word vectorization and Wikipedia's entire corpus to predict the answer to analogies like those above!

This program has around a 19% accuracy rate, which might seem terrible but hear me out: the algorithm is not choosing from a pool of words (like multiple choice); it's actually supplying its own output (like short answer on tests). From past experience, I can say randomly guessing on short answers does NOT give correct solutions 1/5 of the time.

How to have fun with Word Analogizer

0. Make sure your machine is using Python 3!!!
1. Clone this repo
2. Download pip
3. Navigate to the word-analogizer folder and type pip install -r requirements.txt
4. Run runme.py
5. Follow the instructions from Terminal and have fun!

Important note: using the default values, it will take around a minute for the algorithm to train on the Wikipedia dataset. Please be patient, and do not mess with the command line until it prompts you for an analogy to solve.