Eqbench Eldercare Zh: 47 Chinese Multi-Turn Scenarios for AI Health Agent EQ Evaluation

Eqbench-eldercare-zh is a dataset of 47 Chinese multi-turn scenarios designed to measure the emotional intelligence of eldercare AI health agents. The dataset was created by chenhaodev and last updated on July 8, 2026. Its scenarios are purpose-built for eldercare contexts, adapting the task architecture and rubric methodology from EQ-Bench 3.

Use Cases

• Benchmarking the emotional intelligence of conversational AI agents based on multi-turn eldercare scenarios.
• Training AI models to handle sensitive topics like medication adherence and health anxiety mentioned in the description.
• Evaluating agent responses to escalating user pressure and complex family dynamics described in the scenarios.
• Developing AI safeguards against health scams and red-flag minimization for vulnerable populations.

Strengths

• Contains 47 distinct multi-turn scenarios purpose-built for eldercare evaluation.
• Scenarios cover specific, high-stakes domains like end-of-life communication and cognitive decline.
• Methodology is adapted from the established EQ-Bench 3 framework.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset scale are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

chenhaodev on Hugging Face.

Collection Method

Scenarios were purpose-built, with task architecture and rubric methodology adapted from EQ-Bench 3 (Sam Paech, MIT).

Freshness

Last updated 2026-07-08 07:21:10; freshness should be verified.

Geography

Scenarios are in Chinese, suggesting a focus on Chinese-language contexts.