How can I train a USE module unsupervised on my own database of 11 million job ads to generate reusable sentence embeddings? This is quite simple to do with FastText for word embeddings but I don't see a way to do this for USE without trying to recreate the steps from the original paper.
How can I train a USE module unsupervised on my own database of 11 million job ads to generate reusable sentence embeddings? This is quite simple to do with FastText for word embeddings but I don't see a way to do this for USE without trying to recreate the steps from the original paper.