DAHL: Domain-specific Automated Hallucination Evaluation of Long-Form Text through a Benchmark Datas...
DAHL: Domain-specific Automated Hallucination Evaluation of Long-Form Text through a Benchmark Dataset in Biomedicine
About this item
Full title
Author / Creator
Publisher
Ithaca: Cornell University Library, arXiv.org
Journal title
Language
English
Formats
Publication information
Publisher
Ithaca: Cornell University Library, arXiv.org
Subjects
More information
Scope and Contents
Contents
We introduce DAHL, a benchmark dataset and automated evaluation system designed to assess hallucination in long-form text generation, specifically within the biomedical domain. Our benchmark dataset, meticulously curated from biomedical research papers, consists of 8,573 questions across 29 categories. DAHL evaluates fact-conflicting hallucinations...
Alternative Titles
Full title
DAHL: Domain-specific Automated Hallucination Evaluation of Long-Form Text through a Benchmark Dataset in Biomedicine
Authors, Artists and Contributors
Author / Creator
Identifiers
Primary Identifiers
Record Identifier
TN_cdi_proquest_journals_3128886757
Permalink
https://devfeature-collection.sl.nsw.gov.au/record/TN_cdi_proquest_journals_3128886757
Other Identifiers
E-ISSN
2331-8422
