Corrected CBOW Performs as well as Skip-gram
This work addresses a long-standing discrepancy in word embedding performance for researchers and practitioners using CBOW, showing that its perceived inferiority was due to implementation errors.
The authors found that the reported underperformance of CBOW word embeddings compared to Skip-gram was due to faulty negative sampling implementations in popular libraries, not fundamental differences. After correcting a bug in the CBOW gradient update, their CBOW embeddings performed competitively with Skip-gram on various tasks while training significantly faster.
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.