How to build a fixed vocab for lifelong NLP task such as text classification?

Question

I have been trying to build a continual learning NLP algorithm, which rely on the word embedding highly. So I need a fixed vocab to maintain the mapping link between the raw