Biomedical Expert Finding Benchmark: Health Claims Matched to Researcher Publication Profiles

Description

Methods

Data Collection Methodology

We constructed this dataset through a multi-stage pipeline combining automated extraction with expert curation.

Gold-Standard Corpus (15,119 claims). We collected 150,028 health news articles (2018-2024) from reputable sources including Reuters, CNN, BBC, and specialized medical news platforms. We extracted health claims that explicitly cited peer-reviewed research and identified their Digital Object Identifiers (DOIs). Using PubMed's API, we retrieved metadata for the 14,393 unique referenced publications and their 153,147 authors, establishing direct claim-to-expert links.

Silver-Standard Corpus (78,285 claims). We expanded the dataset using four complementary techniques:

1. News article clustering based on temporal proximity, semantic similarity, and named entity overlap to identify additional articles covering the same research.
2. Controlled paraphrasing using language models to increase linguistic diversity.
3. Headline style transfer to simulate different claim formulations.
4. Citation network analysis to identify related publications and their authors.

Expert Profiles. For each researcher, we compiled a comprehensive publication profile by querying PubMed for all their publications from 2018-2024. Expert representations consist of concatenated publication titles and abstracts, capturing their research focus and domain expertise.

Data Format. The dataset includes:

• Claim texts
• Expert identifiers
• Publication metadata (titles, abstracts, publication dates, MeSH terms where available)
• Relevance labels indicating the strength of claim-expert matches
• Clear marking of gold and silver subsets to support different evaluation scenarios

All data were anonymized where appropriate, while preserving the critical relationships between claims and relevant expert publications. The dataset is released to support reproducible research in expert finding and health misinformation detection.

Other

How to Cite

If you use this dataset in your research, please cite both the dataset and the associated paper:

Dataset Citation: 

Zuo, C., Wang, Chenlu, & Banerjee, R. (2025). Biomedical Expert Finding Benchmark: Health Claims Matched to Researcher Publication Profiles [Dataset]. Zenodo. https://doi.org/10.5281/zenodo.15009723

@dataset{biomedicalexpert2025,
  author = {Zuo, Chaoyuan and Wang, Chenlu and Banerjee, Ritwik},
  title = {{Biomedical Expert Finding Benchmark: Health Claims Matched to Researcher Publication Profiles}},
  year = {2025},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.15009723},
  url = {https://doi.org/10.5281/zenodo.15009723}
}

**Paper Citation:**

Zuo, C., Wang, Chenlu, & Banerjee, R. (2025). Large-scale Biomedical Expert Finding for Health Claim Verification: A PubMed-based Retrieval Framework. IEEE International Conference on Bioinformatics and Biomedicine (BIBM).

.
BibTeX:

@inproceedings{healthclaimexperts2025paper,
  author = {Zuo, Chaoyuan and Wang, Chenlu and Banerjee, Ritwik},
  title = {{Large-scale Biomedical Expert Finding for Health Claim Verification: A PubMed-based Retrieval Framework}},
  booktitle = {IEEE International Conference on Bioinformatics and Biomedicine (BIBM)},
  year = {2025}
}

Citation Note: We recommend citing both the dataset and the paper to acknowledge the benchmark creation and methodology. If you use only specific subsets (gold-standard or silver-standard), please mention this in your paper's methodology section.