Generating Long Financial Report Using Conditional Variational Autoencoders With Knowledge Distillation

Generating financial reports from a piece of news is a challenging task due to the lack of sufficient background knowledge to effectively generate long financial reports. To address this issue, this article proposes a conditional variational autoencoders (CVAEs)-based approach that distills external knowledge from a set of news–report data. Specifically, we design an encoder–decoder architecture to learn the latent variable distribution from this set of news–report data to provide background knowledge. Next, a teacher–student network is employed to distill knowledge to refine the output of the decoder component. To evaluate the model performance, extensive experiments have been performed on two public datasets using evaluation criteria like BLEU, ROUGE, METEOR, and human evaluation. Our promising experimental results demonstrate that our proposed approach outperforms existing state-of-the-art approaches.