In the last post we built the data preprocessing required for word2vec training. In this post we will build the network and perform the training on the text8 dataset (source), a Wikipedia dump of ~17 million tokens.

Note that we are implementing the skip-gram version of word2vec since it has superior performance
Continue reading