Antihallucination Dataset: Ground Truth and Generated Text Pairs

A text dataset for analyzing hallucinations in generated language. Each sample contains a ground truth reference document followed by a generated text segment for comparison. The dataset was created by clam004 and last updated on 2024-04-10.

Use Cases

• Train hallucination detection models based on the paired ground truth and generated text structure.
• Benchmark the factual accuracy of text generation models based on the provided reference documents.
• Develop metrics for quantifying hallucination severity based on the comparative text samples.

Strengths

• Explicitly structured with special tokens (<truth> and <generated>) to delineate ground truth from generated content.
• Last updated on 2024-04-10, indicating recent maintenance.

Limitations

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

Provenance

Source

huggingface, derived from the wiki_bio_gpt3_hallucination dataset by potsawee.

Collection Method

Likely contains text pairs structured for comparison, but the exact collection method is not specified.

Freshness

Last updated 2024-04-10 22:53:41