An exploration of mutual information based on emotion–cause pair extraction

摘要

Emotion–cause pair extraction task (ECPE) aims to extract the emotions and causes from an unannotated text. The previous works are mostly limited to using deep networks to model the relation between the emotion clause and cause clause and lack exploration of the statistical dependence between them, such as the effects of emotion–cause causality on the mutual information of two clauses. In this paper, we preliminarily explore the difference between emotion–cause pairs and non emotion–cause pairs in their mutual information and further probe the relations among mutual information, emotion–cause pair, and their relative distance. Additionally, we find that mutual information can be used to measure the dependence strength of an emotion–cause causality on the context. Specifically, we formalize the ECPE as a probability problem and derive the joint distribution of the emotion clause and cause clause using the total probability formula. Based on the joint distribution, we estimate the mutual information (MI) between the emotion clause and cause clause, and further quantify the trend of mutual information in the training process. We conduct various experiments, and the experimental results show that our preliminary exploration is practical and effective. Meanwhile, we also prove our conjecture on the emotion–cause causality and mutual information.