minh and teh icml2012
db2b4182156b2f1f817860ac9f409ad7-Reviews.html
The objective here is to learn word embeddings, that are then evaluated on analogy-based word-similarity tasks. Leveraging the fast training afforded by NCE and using a sligthly simpler model than Minh and Teh ICML2012 they are able to outperform state-of-the-art method of Mikolov et al. 2013, using four times less data and an order of magnitude less computing time.