According to Micholov paper I want to compute Doc2Vec using Keras. I'm new on Keras so I need your help.
There is a corpus of documents with an Id and I want to get two embeddings matrices : one for words and one for paragraphs, isn't it ?
Is it possible to adapt my Word2Vec code to get these embeddings ?
This an extract of my W2V code :
from keras.models import Sequential
cbow = Sequential()
cbow.add(Embedding(input_dim=V, output_dim=dim,input_length=window_size*2))
cbow.add(Lambda(lambda x: K.mean(x, axis=1), output_shape=(dim,)))
cbow.add(Dense(V, activation='softmax'))
Should I add another embedding layer to take into account the paragraph id ?
