The paper presents EMPATH, a new multilingual multi-turn benchmark for safety evaluation of emotional-support chatbots that uses separate auditor and judge models and releases its pipeline and rubrics.
hub
Towards emotional support dialog systems
12 Pith papers cite this work, alongside 189 external citations. Polarity classification is still indexing.
12
Pith papers citing it
189
external citations · OpenAlex
hub tools
years
2026 12representative citing papers
Interviews with AI companion users indicate that virtual embodiment introduces support-intrusion tensions, makes relationships socially legible, and heightens risks of dependence and judgment.
Introduces the SSR dataset and fine-tunes LLMs via chain-of-thought to simulate context-sensitive self-stigma in patient agents based on the 3A1H model.
Introduces a clean matched benchmark and Dynamic Emotional Signature Graphs (DESG) framework that detects implicit sycophancy via clinical-state transitions and reports a 0.0488 macro-F1 gain over baselines on harmful-risk detection.
Rabtriever distills a generative reranker into an efficient bi-encoder using on-policy JEPA to achieve near-reranker accuracy with linear complexity on rationale-based retrieval.
Low-agreeableness persona conditioning in fine-tuning data reduces jailbreak susceptibility and harmful outputs in warm LLMs while preserving conversational warmth.
A hypernetwork produces a condition-dependent beta that meta-gates SwiGLU nonlinearity, giving LLMs adaptive behavior across task, domain, persona and style inputs without finetuning.
SAGE uses a Next Strategy Classifier and Graph-Aware Attention on a psychologically grounded graph to improve LLM strategy prediction and response quality in online counseling.
A multi-agent framework decomposes multimodal empathetic response generation into structured reasoning steps and uses global reflection to reduce emotional biases, outperforming prior methods on IEMOCAP and MELD benchmarks.
MICA mixes per-turn and whole-trajectory normalized reward signals to train emotional-support chatbots, outperforming GRPO and REINFORCE++ on EMPA, EQ-Bench, and EmoBench.
LinguIUTics team applies QLoRA fine-tuning of Qwen3-8B plus stratified CV, minority lexical augmentation, logit bias tuning and ensemble blending to achieve 0.3917 macro F1 (7.7 points above Ministral-8B baseline) on PsyDefDetect 2026.
citing papers explorer
-
EMPATH: A Multilingual Auditor-Judge Benchmark for Safety Evaluation of Emotional-Support Chatbots
The paper presents EMPATH, a new multilingual multi-turn benchmark for safety evaluation of emotional-support chatbots that uses separate auditor and judge models and releases its pipeline and rubrics.
-
"If I Can See You": Understanding Spatially Situated Virtual Embodiment in Close Human-AI Relationships
Interviews with AI companion users indicate that virtual embodiment introduces support-intrusion tensions, makes relationships socially legible, and heightens risks of dependence and judgment.
-
SSR: Can Simulated Patients Learn to Stigmatize Themselves? Modeling Self-Stigma through Internal Monologue
Introduces the SSR dataset and fine-tunes LLMs via chain-of-thought to simulate context-sensitive self-stigma in patient agents based on the 3A1H model.
-
Auditing Stealth Sycophancy in Mental-Health Dialogue: Structured Clinical-State Diagnostics and Clean Matched Benchmarks
Introduces a clean matched benchmark and Dynamic Emotional Signature Graphs (DESG) framework that detects implicit sycophancy via clinical-state transitions and reports a 0.0488 macro-F1 gain over baselines on harmful-risk detection.
-
Efficient Rationale-based Retrieval: On-policy Distillation from Generative Rerankers based on JEPA
Rabtriever distills a generative reranker into an efficient bi-encoder using on-policy JEPA to achieve near-reranker accuracy with linear complexity on rationale-based retrieval.
-
Low-Agreeableness Persona Conditioning for Safe LLM Fine-Tuning
Low-agreeableness persona conditioning in fine-tuning data reduces jailbreak susceptibility and harmful outputs in warm LLMs while preserving conversational warmth.
-
Learn-To-Learn on Arbitrary Textual Conditioning: A Hypernetwork-Driven Meta-Gated LLM
A hypernetwork produces a condition-dependent beta that meta-gates SwiGLU nonlinearity, giving LLMs adaptive behavior across task, domain, persona and style inputs without finetuning.
-
SAGE: A Strategy-Aware Graph-Enhanced Generation Framework For Online Counseling
SAGE uses a Next Strategy Classifier and Graph-Aware Attention on a psychologically grounded graph to improve LLM strategy prediction and response quality in online counseling.
-
A Multi-Agent Framework with Structured Reasoning and Reflective Refinement for Multimodal Empathetic Response Generation
A multi-agent framework decomposes multimodal empathetic response generation into structured reasoning steps and uses global reflection to reduce emotional biases, outperforming prior methods on IEMOCAP and MELD benchmarks.
-
MICA: Multi-granularity Intertemporal Credit Assignment for Long-Horizon Emotional Support Dialogue
MICA mixes per-turn and whole-trajectory normalized reward signals to train emotional-support chatbots, outperforming GRPO and REINFORCE++ on EMPA, EQ-Bench, and EmoBench.
-
LinguIUTics at PsyDefDetect: Iterative Imbalance-Aware Fine-tuning of Qwen3-8B for Psychological Defense Mechanism Classification
LinguIUTics team applies QLoRA fine-tuning of Qwen3-8B plus stratified CV, minority lexical augmentation, logit bias tuning and ensemble blending to achieve 0.3917 macro F1 (7.7 points above Ministral-8B baseline) on PsyDefDetect 2026.
- VISHC at PsyDefDetect: Mitigating Data Scarcity in Psychological Defense Classification with Context-Aware Synthetic Augmentation