Contrastive Learning for Neural Topic Model

Nonetheless, this framework has two main limitations. First, A TM relies on the key ingredient: leveraging the discrimination of the real distribution from the fake (negative) distribution to guide the training.