Corrected CBOW Performs as well as Skip-gram
Abstract
Mikolov et al. (2013a) observed that continuous bag-of-words (CBOW) word embeddings tend to underperform Skip-gram (SG) embeddings, and this finding has been reported in subsequent works. We find that these observations are driven not by fundamental differences in their training objectives, but more likely on faulty negative sampling CBOW implementations in popular libraries such as the official implementation, word2vec.c, and Gensim. We show that after correcting a bug in the CBOW gradient update, one can learn CBOW word embeddings that are fully competitive with SG on various intrinsic and extrinsic tasks, while being many times faster to train.
Cite
@article{arxiv.2012.15332,
title = {Corrected CBOW Performs as well as Skip-gram},
author = {Ozan İrsoy and Adrian Benton and Karl Stratos},
journal= {arXiv preprint arXiv:2012.15332},
year = {2021}
}
Comments
Presented at WINR at EMNLP 2021, added discussion about FastText, more discussion about findings, additional results on C4 data, wording changes