Word2Vec Loss Function

[sidana...@gmail.com]

Hi,

I am using the following code to implement item2vec:

import gensim

from gensim.models import Word2Vec

from gensim.models.word2vec import LineSentence

sentences = LineSentence('items')

model = Word2Vec(sentences, size=20, window=3, min_count=0, workers=4,sg=1)

It works perfectly.

But, I have difficulty in understanding the loss function (objective) which word2vec minimizes or maximizes while coming up with embeddings. Can anyone please shed light on what is the optimization problem being solved inside Gensim for word2vec.

Thanks,

Sumit

[Ivan Menshikh]

Hello,

Look at section 2 ("The Skip-gram Model")  https://arxiv.org/abs/1310.4546