AutoCBT: An Adaptive Multi-Agent Framework for Enhanced Automated Cognitive Behavioral Therapy

Traditional counseling, which is often done in person, is often limited to people seeking help with psychological problems. In contrast, online spontaneous counseling offers a viable option for those who are reluctant to pursue treatment due to stigma or shame. Cognitive Behavioral Therapy (CBT), a widely used method of counseling, aims to help people identify and correct cognitive distortions that contribute to negative emotions and behaviors. The emergence of LLMs has opened up new possibilities for automating the diagnosis and treatment of CBT. However, current LLM-based CBT programs face challenges such as fixed structural frameworks, which limit adaptation and personalization, and repetitive response patterns that provide generic, unhelpful suggestions.

Recent advances in AI have introduced frameworks such as CBT-LLM, which uses fast-based learning, and CoCoA, which combines memory methods to create incremental-retrieval. These programs aim to identify and address cognitive distortions in user statements while improving the depth and validity of the therapeutic interaction. Despite their strengths, existing methods often lack personalization, flexibility to user needs, and little understanding of adaptive treatment processes. To close these gaps, ongoing research uses annotated datasets, ontologies, and advanced LLMs to develop context-aware CBT systems that mimic human cognitive processes.

Researchers from Shenzhen Key Laboratory for High-Performance Data Mining, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, and several other institutions have developed AutoCBT, an independent multi-agent framework for CBT in one-on-one counseling . Using models like Quora and YiXinLi, AutoCBT combines dynamic routing and memory methods to improve response quality and adaptability. The framework addresses systematic thinking and planning to produce high-quality, context-aware results. Tested on a bilingual dataset, it outperforms conventional LLM-based systems, addressing challenges such as dynamic routes, routing paths, and Llama's overprotection problem.

AutoCBT is a flexible framework designed for multi-agent systems in CBT, including Counselor Agent (interface), Supervisor Agents, communication topology, and routing strategies. Agent Advisor, powered by LLMs, interacts with users and seeks input from Executives to generate confident, high-quality responses. Agents include short- and long-term memory mechanisms, and routing techniques such as unicast and broadcast enable dynamic communication. AutoCBT combines the principles of CBT—empathy, identification of beliefs, reflection, strategy, and encouragement—mapped to specific Supervisor Agents. Its performance was validated using a bilingual dataset including PsyQA and TherapistQA, segmented and supplemented with examples of cognitive distortions.

In online counseling, LLMs such as Qwen-2.5-72B and Llama-3.1-70B have been tested to handle emotional nuances and follow instructions. AutoCBT, a two-stage framework, Generation and PromptCBT works well by combining flexible routing and guidance methods, achieving high scores across empathy, managing cognitive distortions, and matching responses. AutoCBT's iterative approach developed its own draft responses, validated by automation and human evaluation. Challenges included route conflicts, role confusion, and unnecessary feedback loops, which were reduced through design refinements. Llama's excessive vigilance led to constant rejection of sensitive topics, in contrast to Qwen, who responded in depth, highlighting the importance of moderation in model sensitivity.

In conclusion, AutoCBT is a new multi-agent framework for CBT-based counseling. By combining flexible routing and guidance methods, AutoCBT addresses the limitations of traditional LLM-based counseling, greatly improving the quality of response and efficiency in identifying and dealing with cognitive distortions. AutoCBT achieves a higher interview quality with its flexible and independent design compared to static, fast-based programs. Challenges in LLM semantic understanding and instructional compliance were identified and mitigated with targeted solutions. Using bilingual datasets and models, the framework demonstrates its potential to deliver high-quality, automated counseling services. It offers a dangerous alternative to people who hesitate to pursue traditional medicine because of stigma.

Check it out Paper. All credit for this study goes to the researchers of this project. Also, don't forget to follow us Twitter and join our Telephone station again LinkedIn Grup. Don't forget to join our 65k+ ML SubReddit.

🚨 [Recommended Read] Nebius AI Studio extends with vision models, new language models, embeddings and LoRA ^(Promoted)

Sana Hassan, a consulting intern at Marktechpost and a dual graduate student at IIT Madras, is passionate about using technology and AI to address real-world challenges. With a deep interest in solving real-world problems, he brings a fresh perspective to the intersection of AI and real-life solutions.