Uncertainty-Aware Abstractive Summarization

We propose a novel approach to summarization based on Bayesian deep learning. We approximate Bayesian summary generation by first extending state-of-the-art summarization models with Monte Carlo dropout and then using them to perform multiple stochastic forward passes. This method allows us to improve summarization performance by simply using the median of multiple stochastic summaries. We show that our variational equivalents of BART and PEGASUS can outperform their deterministic counterparts on multiple benchmark datasets. In addition, we rely on Bayesian inference to measure the uncertainty of the model when generating summaries. Having a reliable uncertainty measure, we can improve the experience of the end user by filtering out generated summaries of high uncertainty. Furthermore, our proposed metric could be used as a criterion for selecting samples for annotation, and can be paired nicely with active learning and human-in-the-loop approaches.