| --- |
| license: mit |
| task_categories: |
| - text-classification |
| - fact-checking |
| language: |
| - en |
| tags: |
| - citation-verification |
| - hallucination-detection |
| - scientific-integrity |
| - bayesian-scoring |
| - reference-checking |
| size_categories: |
| - 10K<n<100K |
| --- |
| |
| # IntegriRef-Bench |
|
|
| Benchmark dataset for multi-level reference integrity verification, accompanying the IntegriRef framework. |
|
|
| ## Dataset Splits |
|
|
| | Split | Rows | Description | |
| |-------|------|-------------| |
| | `reference_verification` | 1,926 | Golden benchmark + crawled verification cases (hallucinated, real, chimera, retracted) | |
| | `signal_unit_tests` | 403 | Per-signal unit tests for 14 Bayesian signal types | |
| | `l1_intent` | 20 | Citation intent classification test pairs | |
| | `l2_nli` | 20 | Claim-evidence NLI alignment test pairs | |
| | `graph_anomaly` | 3,030 | Citation graph anomaly cases (rings, temporal, orphan clusters) | |
| | `retracted_papers` | 6,391 | Retracted papers from Crossref + PubMed with real controls | |
| | **Total** | **11,790** | | |
|
|
| ## Usage |
|
|
| ```python |
| from datasets import load_dataset |
| |
| ds = load_dataset("Geoffrey-Wang/IntegriRef-Bench") |
| |
| # Or load a specific split |
| ref_ver = load_dataset("Geoffrey-Wang/IntegriRef-Bench", data_files="reference_verification.jsonl") |
| ``` |
|
|
| ## Sources |
|
|
| - **Retracted papers**: Crossref `update-to` retraction markers + PubMed retraction notices |
| - **Hallucinated references**: Programmatically generated with verified non-existence |
| - **Chimera references**: Real DOIs paired with swapped metadata |
| - **Graph anomaly cases**: Documented citation cartels (Brazilian 2009-2013, Ji-Huan He, IOP 2024) |
| - **Temporal anomalies**: OpenAlex citation graph analysis |
| - **Statistical cases**: GRIM test + statcheck from PMC Open Access full texts |
|
|
| ## Citation |
|
|
| ```bibtex |
| @inproceedings{integriref2026, |
| title={IntegriRef: A Five-Layer Bayesian Framework for Cross-Domain Reference Integrity Verification}, |
| author={Anonymous}, |
| booktitle={KnowFM Workshop at ACL}, |
| year={2026} |
| } |
| ``` |
|
|
| ## License |
|
|
| MIT |
|
|
