Event Temporal Relation Extraction with Bayesian Translational Model

TL;DR

This paper introduces Bayesian-Trans, a probabilistic model for event temporal relation extraction that leverages Bayesian inference to better encode uncertainty and improve accuracy over traditional neural methods.

Contribution

The paper presents Bayesian-Trans, a novel Bayesian learning approach that models temporal relations as latent variables and infers their distributions, outperforming existing methods.

Findings

Bayesian-Trans outperforms existing approaches on three datasets.

The model effectively quantifies uncertainty in predictions.

Detailed analyses demonstrate the benefits of Bayesian inference in this task.

Abstract

Existing models to extract temporal relations between events lack a principled method to incorporate external knowledge. In this study, we introduce Bayesian-Trans, a Bayesian learning-based method that models the temporal relation representations as latent variables and infers their values via Bayesian inference and translational functions. Compared to conventional neural approaches, instead of performing point estimation to find the best set parameters, the proposed model infers the parameters' posterior distribution directly, enhancing the model's capability to encode and express uncertainty about the predictions. Experimental results on the three widely used datasets show that Bayesian-Trans outperforms existing approaches for event temporal relation extraction. We additionally present detailed analyses on uncertainty quantification, comparison of priors, and ablation studies, illustrating the benefits of the proposed approach.