Following this paper and it's code, we added a zero-shot learning approach where our model could predict words even out of training data, by making the word predict word embedding instead of one-hot vector.
You can find the detailed report here
Following this paper and it's code, we added a zero-shot learning approach where our model could predict words even out of training data, by making the word predict word embedding instead of one-hot vector.
You can find the detailed report here