This repository contains a Runway model for experimenting with Stanford NLP's GloVe, an unsupervised learning algorithm for obtaining vector representations for words. Similar to Word2Vec, GloVe creates a continuous N-dimensional representation of a word that is learned from its surrounding context words in a training corpus. Trained on a large corpus of text, these co-occurance statistics (an N-dimensional vector embedding) cause semantically similar words to appear near each-other in their resulting N-dimensional embedding space (e.g. "dog" and "cat" may appear nearby a region of other pet related words in the embedding space because the context words that surround both "dog" and "cat" in the training corpus are similar).

This repo is a hard-fork of the brannondorsey/GloVe-experiments repo with Runway-specific code changes and optimizations using FAISS.

All code is released under an MIT license. You are free to copy, edit, share, or sell it under those terms.

Jeffrey Pennington, Richard Socher, and Christopher D. Manning. 2014. GloVe: Global Vectors for Word Representation.

@inproceedings{pennington2014glove,
  author = {Jeffrey Pennington and Richard Socher and Christopher D. Manning},
  booktitle = {Empirical Methods in Natural Language Processing (EMNLP)},
  title = {GloVe: Global Vectors for Word Representation},
  year = {2014},
  pages = {1532--1543},
  url = {https://www.aclweb.org/anthology/D14-1162},
}