RevUp: Revise and Update Information Bottleneck for Event Representation

TL;DR

This paper introduces RevUp, a semi-supervised discrete latent variable model that effectively incorporates noisy or missing external knowledge into event representations, outperforming previous methods.

Contribution

RevUp proposes a novel hierarchical reparameterization and mutual information minimization approach for better event modeling with side knowledge.

Findings

Outperforms previous approaches on multiple datasets

Theoretically generalizes past methods

Effective handling of noisy or missing side information

Abstract

The existence of external (``side'') semantic knowledge has been shown to result in more expressive computational event models. To enable the use of side information that may be noisy or missing, we propose a semi-supervised information bottleneck-based discrete latent variable model. We reparameterize the model's discrete variables with auxiliary continuous latent variables and a light-weight hierarchical structure. Our model is learned to minimize the mutual information between the observed data and optional side knowledge that is not already captured by the new, auxiliary variables. We theoretically show that our approach generalizes past approaches, and perform an empirical case study of our approach on event modeling. We corroborate our theoretical results with strong empirical experiments, showing that the proposed method outperforms previous proposed approaches on multiple datasets.