Was very pleased to find this out when I first started studying word embeddings (the abstract principles of word2vec). Essentially it comes down to words having similar verbs and objects that come up most frequently together, so they end up being semantically close.
