Rationale
Although continuous bag of word (CBOW) embeddings can be trained more quickly than skipgram (SG) embeddings, it is a common belief that SG embeddings tend to perform better in practice. This was observed by the original authors of Word2Vec [1] and also in subsequent work [2]. However, we found that popular implementations of word2vec with negative sampling such as word2vec and gensim do not implement the CBOW update correctly, thus potentially leading to misconceptions about the performance of CBOW embeddings when trained correctly.
We release kōan so that others can efficiently train CBOW embeddings using the corrected weight update. See this technical report for benchmarks of kōan vs. gensim word2vec negative sampling implementations. If you use kōan to learn word embeddings for your own work, please cite: