CARE-XAI: Culturally-Aware Health Claim Verification Dataset

CARE-XAI consolidates five public health NLP benchmarks into a unified schema for verifying health claims. The dataset contains 17,803 total rows, split into training, validation, and test sets. It was created by Prabhjotschugh and last updated on Hugging Face in May 2026.

Use Cases

• Training models for health claim verification based on consolidated public benchmarks.
• Evaluating the cultural relevance of AI explanations for health information based on cultural relevance flags.
• Assessing the quality of evidence supporting health claims based on GRADE evidence quality labels.
• Benchmarking explanation generation systems using the provided Gold/Silver explanation annotations.

Strengths

• Unifies five distinct public health NLP benchmarks (17,803 rows) into a single schema.
• Includes structured annotations for evidence quality (GRADE) and cultural relevance.
• Provides a standard train/validation/test split (14,254/1,797/1,752 rows).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Consolidated from five public health NLP benchmarks, including PUBHEALTH.

Collection Method

Dataset consolidation and annotation augmentation.

Freshness

Last updated 2026-05-10 09:28:40; freshness should be verified.