Causal Discovery Inspired Unsupervised Domain Adaptation for Emotion-Cause Pair Extraction

2024-06-18

Conference on Empirical Methods in Natural Language Processing

AI Summary

This work addresses the unsupervised domain adaptation challenge in cross-domain emotion-cause pair extraction, where target-domain event distributions exhibit severe shift while emotion expression distributions remain relatively stable. To mitigate spurious correlations induced by source-domain events, we propose— for the first time in this task—a causal-discovery-inspired framework based on variational autoencoders (VAEs) that explicitly disentangles latent representations of emotions and events via variational posterior regularization. This decoupling enhances generalization to unseen target-domain events without supervision. Our method achieves substantial improvements over state-of-the-art unsupervised adaptation approaches, boosting weighted F1 scores by 11.05% on Chinese benchmarks and 2.45% on English benchmarks. To foster reproducibility and further research, we release both the source code and the generated dataset.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Adversarial Attacks & Robustness
• Reasoning under Uncertainty: Causality

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Web Mining and Content Analysis: Large pretrained models with web data
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

This paper tackles the task of emotion-cause pair extraction in the unsupervised domain adaptation setting. The problem is challenging as the distributions of the events causing emotions in target domains are dramatically different than those in source domains, despite the distributions of emotional expressions between domains are overlapped. Inspired by causal discovery, we propose a novel deep latent model in the variational autoencoder (VAE) framework, which not only captures the underlying latent structures of data but also utilizes the easily transferable knowledge of emotions as the bridge to link the distributions of events in different domains. To facilitate knowledge transfer across domains, we also propose a novel variational posterior regularization technique to disentangle the latent representations of emotions from those of events in order to mitigate the damage caused by the spurious correlations related to the events in source domains. Through extensive experiments, we demonstrate that our model outperforms the strongest baseline by approximately 11.05% on a Chinese benchmark and 2.45% on a English benchmark in terms of weighted-average F1 score. We have released our source code and the generated dataset publicly at: https://github.com/tk1363704/CAREL-VAE.

Problem

• Unsupervised domain adaptation for emotion-cause pairs.
• Different event distributions across source and target domains.
• Mitigate spurious correlations in source domain events.

Innovation

• VAE framework for latent structures
• Variational posterior regularization technique
• Disentangle emotion and event representations