Working on genre classification for the movielens dataset. Apart from the movie description, the dataset consists of other features as well, like, cast, tagline, etc. I am wanting to use ULMFit to perform the text classification. What I am wanting to do is to add an embedding layer for each of these features. But being new to all this, I am a little unsure how to go forward. The AWD LSTM used in fastai has the encoder like this:
self.encoder = nn.Embedding(vocab_sz, emb_sz, padding_idx=pad_token)
Now, for each feature, the embedding layer will have a different
vocab_sz . Do I need to concatenate them all and sum up the different vocab sizes to form a new embedding layer? Is that meaningful? If not, what is the best way to incorporate such features?