CausalESC: Breaking Causal Cycles for Emotional Support Conversations with Temporal Causal HMM

20 Sept 2024 (modified: 25 Nov 2024)ICLR 2025 Conference Withdrawn SubmissionEveryoneRevisionsBibTeXCC BY 4.0
Keywords: Emotional Support Conversation, Causal Learning, Text Generation
TL;DR: CausalESC breaks causal cycles and dynamically captures the dynamic evolution of the seeker's cognition, behavior, and emotion for generating more effective supportive responses.
Abstract: Emotional Support Conversation (ESC) is a rapidly advancing task focused on alleviating a seeker's emotional distress. The intricate interplay between cognition, emotion, and behavior presents substantial challenges for existing approaches, which often struggle to capture the dynamic evolution of the seeker's internal state during conversations. To address this, we propose \textbf{CausalESC}, a model designed to dynamically represent the seeker's internal states, by assuming that the generative process governing the mutual influence among these factors follows a first-order Markov property, with \iid random variables. The model comprises a prior network, that disentangles the seeker's emotions, cognition, and behavior, and a posterior network, which decouples the support strategy factors. The prior network also models the psychological causality of the seeker within each conversation round. To account for the varying effects of support strategies on the seeker's intrinsic states, we incorporate a support intervention module to capture these impacts. Additionally, a holistic damping transfer mechanism is designed to regulate the complex interactions among cognition, emotion, behavior, and strategy, ensuring that changes remain within a reasonable range. Our model effectively breaks causal cycles and achieves causal representation learning. Both automatic and human evaluations demonstrate the effectiveness of our model, emphasizing the advantages of modeling the evolution of the seeker's internal state under support strategies.
Primary Area: applications to computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Supplementary Material: zip
Submission Number: 2111
Loading