A simple spell checker built from word vectors – Ed Rushton – Medium

Computers only deal with numbers, and so to get a computer to analyse text data -- for example, to find topics, to translate, to summarise, etc -- you must first convert the data into numbers. A'word vector' is simply a set of numbers which represent a word: the computer's internal representation of that word. If we train a computer to predict the missing word from a sentence, giving it millions of examples to learn from, and we allow the computer to improve its predictions by changing the numbers allocated to each word, we find that synonyms end up being allocated numbers that are close to one another. There are lots of blog posts and tutorials out there which explain the mechanics behind the word vector training process. My aim below is to give an understanding of why the words end up in the places they do -- of why synonyms end up close together.