Graph2Counsel: Clinically Grounded Synthetic Counseling Dialogue Generation from Client Psychological Graphs

Aishik Mandal, Hiba Arnaout, Clarissa W. Ong, Juliet Bockhorst, Kate Sheehan, Rachael Moldow, Tanmoy Chakraborty, Iryna Gurevych, Technische Universität Darmstadt Zuse School ELIZA · 2026

Read Graph2Counsel: Clinically Grounded Synthetic Counseling Dialogue Generation from Client Psychological Graphs by Aishik Mandal, Hiba Arnaout, Clarissa W. Ong, Juliet Bockhorst, Kate Sheehan, Rachael Moldow, Tanmoy Chakraborty, Iryna Gurevych, Technische Universität Darmstadt Zuse School ELIZA (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-20382.txt.

Introduction

Mandal and seven colleagues present Graph2Counsel, a frame for building synthetic counseling sessions from client psychological graphs. The graphs encode links among client thoughts, feelings, and behaviors, and a prompting pipeline guided by counselor strategies turns graphs into sessions. Prompting styles include chain-of-thought and multi-agent feedback, yielding 760 sessions from 76 graphs across client profiles. Expert review rates the set above past sets on specificity, counselor skill, authenticity, flow, and safety, with agreement near 0.70, and fine-tuning lifts CounselingBench and CounselBench scores.

Catalog record

CREATORAishik Mandal, Hiba Arnaout, Clarissa W. Ong, Juliet Bockhorst, Kate Sheehan, Rachael Moldow, Tanmoy Chakraborty, Iryna Gurevych, Technische Universität Darmstadt Zuse School ELIZA
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE134644
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.20382

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Graph2Counsel: Clinically Grounded Synthetic Counseling Dialogue Generation from Client Psychological Graphs Aishik Mandal, Hiba Arnaout, Clarissa W. Ong, Juliet Bockhorst, Kate Sheehan, Rachael Moldow, Tanmoy Chakraborty, Iryna Gurevych, Technische Universität Darmstadt Zuse School ELIZA Abstract Rising demand for mental health support has increased interest in using Large Language Models (LLMs) for counseling. However, adapting LLMs to this high-risk safety-critical domain is hindered by the scarcity of real-world counseling data due to privacy constraints. Synthetic datasets provide a promising alternative, but existing approaches often rely on unstructured or semi-structured text inputs and overlook structural dependencies between a client’s cognitive, emotional, and behavioral states, often producing psychologically inconsistent interactions and reducing data realism and quality. We introduce Graph2Counsel, a framework for generating synthetic counseling sessions grounded in Client Psychological Graphs (CPGs) that encode relationships among clients’ thoughts, emotions, and behaviors. Graph2Counsel employs a structured prompting pipeline guided by counselor strategies and CPG, and explores prompting strategies including CoT 52 and Multi-Agent Feedback 26 . Graph2Counsel produces 760 sessions from 76 CPGs across diverse client profiles. In expert evaluation, our dataset outperforms prior datasets on specificity, counselor competence, authenticity, conversational flow, and safety, with substantial inter-annotator agreement (Krippendorff’s α \alpha = 0.70). Fine-tuning an open-source model on this dataset improves performance on CounselingBench 40 and CounselBench 27 , showing downstream utility. We also make our code and data public. 1 1 1 Graph2Counsel code and data 1 Introduction Mental health disorders affect one in seven people worldwide 2 2 2 WHO (2025): https://www.who.int/news-room/fact-sheets/detail/mental-disorders , yet access to counseling remains limited due to clinician shortages, cost, and stigma. Consequently, many individuals turn to AI systems for mental health support because they are accessible and non-judgmental. However, deploying general-purpose Large Language Models (LLMs) without adaptation poses risks, including hallucinated advice, misaligned interventions, and potential psychological harm 11 . Addressing these risks requires high-quality counseling dialogue data, but collecting large-scale datasets from real counseling sessions is challenging due to strict confidentiality and ethical constraints 34 . Even when transcripts are available, anonymization methods such as manual de-identification or automatic pseudonymization 49 ; 58 remain limited in scalability and robustness. Consequently, publicly available counseling datasets remain limited, motivating synthetic session generation to expand counseling data without exposing sensitive client information. Figure 1: From a real therapy transcript, we extract a Client Psychological Graph (CPG) 41 , whose nodes represent psychological processes (e.g., fear of judgment) and edges capture their relationships (e.g., excites or inhibits). From the same transcript, we also extract counselor strategies (e.g., reframing, empathy building). We then use the CPG to generate diverse client profiles, and finally combine the profile, CPG, and counselor strategies to generate synthetic counseling dialogues. Early counseling datasets such as Psych8k 29 and MentalChat16k 54 contain single-turn question–answer pairs from real sessions. Multi-turn datasets, including MDD-5k 57 and MusPsy 51 , extend this setting using client profiles derived from real sessions. Other work adds psychological grounding through counselor notes in CPsyCoun 59 and symptom lists in 9 . PsyDial 47 generates masked client utterances from real sessions but does not model the clinical reasoning underlying therapy. More recent approaches such as CACTUS 24 , MAGneT 35 , and SQPsychConv 50 introduce structured client profiles or questionnaire-derived descriptions. However, most synthetic counseling datasets still rely on static or unstructured inputs (e.g., demographics, symptoms, or questionnaire scores) that capture only isolated snapshots of a client’s state and overlook dependencies among cognitive, emotional, and behavioral processes, e.g., indicating overthinking and fear of judgment without modeling the dependency between these factors. Addressing these limitations requires structured representations that encode interactions among cognitive, emotional, and behavioral states. Client Psychological Graphs (CPGs) 4 ; 15 ; 25 ; 42 ; 41 provide such a representation: nodes denote psychological processes (e.g., fear of judgment), while directed edges capture functional relations (e.g., fear of judgment triggering overthinking, or mindfulness reducing it). Grounding dialogue generation in CPGs enables models to capture reasoning and emotional dynamics overlooked by text-centric methods. Recent work 41 shows that CPGs can be extracted from counseling transcripts using prompt-based LLM pipelines. Unlike transcripts, CPGs encode relational dynamics without sensitive personal content, offering a compact and interpretable representation of psychological processes. Based on this representation, we introduce Graph2Counsel (Figure 1 ), a framework for generating CPG-grounded synthetic counseling sessions. By conditioning dialogue generation on CPGs and incorporating counselor strategies (interventions), Graph2Counsel produces multi-turn dialogues that reflect clinically meaningful interactions among psychological states. Table 1 provides a concise comparison of Graph2Counsel with existing synthetic counseling datasets. Our main contributions are summarized below: Method Therapy Modality (Type) Input Contextual Modeling Avg. Turns Size Expert Eval. Psych8k ( 29 ) Unspecified Real counseling dialogues text-based 2.00 8,187 ✗ MDD-5k ( 57 ) Diagnosis Real client profiles text-based 53.60 5,000 ✓ CPsyCoun ( 59 ) Unspecified Online counseling dialogues text-based 17.40 3,134 ✗ CACTUS ( 24 ) Cognitive behavioral therapy Crowdsourced simulated client profiles text-based 33.20 31,577 ✓ MAGneT ( 35 ) Cognitive behavioral therapy Crowdsourced simulated client profiles text-based 42.00 442 ✓ SQPsychConv ( 50 ) Cognitive behavioral therapy Structured questionnaires from real clients text-based 33.98 2090 ✓ Graph2Counsel ( ours ) Process-based (meta-framework) CPGs from real sessions and graph-based client profiles text- + graph-based 40.12 760 ✓ Table 1: Comparison with related synthetic counseling datasets. Process-based therapy, a meta-framework integrating multiple therapy modalities (different types of therapy), reveals 29 modalities in Graph2Counsel (Appendix D ). Our dynamic contextual modeling captures Thought–Emotion–Behavior interactions via Client Psychological Graphs (CPGs). 1. We introduce Graph2Counsel , a framework for generating CPG-grounded synthetic counseling sessions. 2. We construct a dataset of 760 synthetic counseling sessions covering 29 therapy modalities, providing a new benchmark for mental health LLMs. 3. We perform ablations over different input representations (CPGs, CPG-derived client profiles, and their combinations) and prompting strategies (Guided Counseling, CoT, Multi-Agent) to analyze their impact on dialogue generation quality, evaluated using the Cognitive Therapy Rating Scale (CTRS) 16 and Working Alliance Inventory (WAI) 22 . 4. We fine-tune Llama3-8B-Instruct ( 37 ) on our dataset and evaluate it on CounselingBench 40 and CounselBench 27 . 5. We conduct expert evaluation with four licensed clinicians, where Graph2Counsel ranks highest compared to existing datasets on counselor competence, authenticity, specificity, conversational flow and safety, with strong inter-annotator agreement (Krippendorff’s \alpha=0.70 ). 2 Related Work Synthetic counseling data generation. General LLMs struggle on counseling tasks 18 , largely due to the scarcity of high-quality data constrained by privacy concerns 34 , motivating synthetic data generation. Early datasets such as Psych8k ( 29 ) and MentalChat16k ( 54 ) focus on single-turn interactions, while later approaches generate multi-turn dialogues using client profiles ( 57 ; 51 ) , counselor notes ( 59 ) , or symptom lists ( 9 ) . These datasets provide basic client attributes (e.g., demographic information, symptoms or presenting problems) but lack functional dependencies among cognitive, emotional, and behavioral processes critical for realistic dialogues. Other approaches generate dialogues directly via LLM prompts ( 5 ) or by first synthesizing client profiles ( 61 ) or notes ( 32 ) . However, without explicit psychological grounding, models often struggle to select clinically appropriate interventions ( 24 ) . HealME ( 53 ) addresses this by using fixed strategies,

Read the complete text