Multi-Phase Counseling Response Generation with Active Cognitive Distortion Recognition

ACL ARR 2025 May Submission6240 Authors

20 May 2025 (modified: 03 Jul 2025)ACL ARR 2025 May SubmissionEveryoneRevisionsBibTeXCC BY 4.0
Abstract: In recent years, there has been great progress in dialogue generation for AI-driven psychological counseling by leveraging Cognitive Behavioral Therapy (CBT). However, existing systems focus largely on short-term exchanges, without mechanisms to model the sequential structure of CBT interventions or to recognize cognitive distortions. These limitations impede their ability to support sustained therapeutic engagement. To address this deficiency, we introduce the Multi-Phase Counseling Dialogues dataset, encompassing simulated therapist-client interactions across three distinct therapeutic phases. The dataset's clinical validity is rigorously assessed utilizing the Cognitive Therapy Rating Scale. After dataset construction, we propose MPAC, a multi-phase counseling dialogue generation framework integrating proactive cognitive distortion recognition. MPAC leverages a fine-tuned module to identify and track clients’ cognitive distortions, alongside a phase-aware mechanism that infers the current therapeutic phase and generates structured, CBT-aligned responses. Empirical results demonstrate that MPAC significantly surpasses state-of-the-art baselines in both the application of CBT techniques and overall counseling efficacy. These findings indicate that combining proactive cognitive-distortion detection with explicit multi-phase modeling can substantially enhance the depth and continuity of mediated psychological interventions.
Paper Type: Long
Research Area: Linguistic theories, Cognitive Modeling and Psycholinguistics
Research Area Keywords: computational psycholinguistics
Contribution Types: Publicly available software and/or pre-trained models, Data resources
Languages Studied: English; Chinese
Submission Number: 6240
Loading