Skip to main content

Frequency effects in large word embeddings

Continuing my explorations of frequency effects in word embeddings, I have a another short series in which I provide some summary tests for the frequency effects, and then show those tests in action, especially on large pre-trained word embeddings: Google News, GLoVe, and FastText.

There are still lots of questions to explore. Next up: I intend to address the question of whether these frequency effects make a difference? In other words, would reducing the effects improve the quality of the embeddings?