The common model for learning word embedding is based on BOW, and Skip-gram model.

Is it possible to train a RNN-based architecture like GRU or LSTM with random sentences from a large corpus to learn word embeddings. Basically, train a network with positive and negative samples, and back-propagate into word vectors. What are the drawbacks of this technique?

Any pointers to similar work is highly appreciated.

The common way of learning word embeddings is based on BOW, and Skip-gram models.

Is it possible to train a RNN-based architecture like GRU or LSTM with random sentences from a large corpus to learn word embeddings? Basically, we train a network with positive and negative samples, and back-propagate into word vectors. What are the drawbacks of this technique?

Any reference to similar works is highly appreciated.