Multi-turn Empathetic Dialogue Benchmark with User Profiles

KardiaBench is a large-scale, user-profile-grounded benchmark for evaluating empathetic dialogue systems. The dataset was created by Jhcircle and was accepted for presentation at WWW 2026, with a last update recorded in March 2026.

Use Cases

• Training reinforcement learning agents for emotional support using a rubric-as-judge method.
• Evaluating model performance on multi-turn empathetic dialogue grounded in user profiles.
• Benchmarking LLM reasoning capabilities toward understanding and empathy in structured conversations.

Strengths

• Dataset is described as large-scale, suggesting a substantial collection of dialogues.
• Features structured reasoning and is grounded in user profiles for multi-turn conversations.

Limitations

• Specific row count, column names, and sample size are unknown, limiting reproducibility.
• The dataset's geographic and demographic coverage is unspecified, which may indicate bias.

Provenance

Source

Jhcircle on Hugging Face.

Collection Method

Designed as a user-grounded, multi-turn empathetic dialogue benchmark; specific collection method is null.

Freshness

Last updated on 2026-03-02.

Geography

Region tag indicates 'us', but specific spatial coverage is null.