CitationHub

Dataset Description

CitationHub is a large-scale scholarly citation dataset designed for citation analysis, citation intent modeling, scholarly knowledge graph construction, and evidence-grounded hypothesis generation for AI co-scientist systems.

The dataset contains citation events between citing papers and highly cited seed papers, enriched with citation intent labels, citation contexts, field information, journal metadata, author and affiliation information, and heterogeneous knowledge graph nodes and edges.

The dataset is intended to support research on:

• Citation intent analysis
• Scientific impact modeling
• Future citation prediction
• Scholarly knowledge graph completion
• Intent-aware citation recommendation
• Evidence-grounded hypothesis generation
• AI co-scientist systems for scientific discovery

Dataset Repository: https://zenodo.org/records/18536895

---

Dataset Summary

File	Rows	Columns	Description
citation_events.parquet	1,857,503	20	Raw citation event records
citation_events_enriched.parquet	1,857,503	32	Citation events enriched with cited seed paper metadata
citation_events_normalized.parquet	1,857,503	23	Citation events with normalized intent and field identifiers
citing_papers.parquet	1,467,045	7	Metadata of citing papers
citing_papers_normalized.parquet	1,467,045	8	Citing papers with normalized journal IDs
seed_cited_papers.parquet	23,479	42	Highly cited seed paper metadata
seed_cited_papers_normalized.parquet	23,479	48	Seed papers with normalized journal, author, affiliation, city, country, and field IDs
kg_edges.parquet	6,855,117	3	Heterogeneous knowledge graph edges
kg_nodes.parquet	3,418,433	14	Heterogeneous knowledge graph nodes
authors.parquet	16,839	2	Author ID-name mapping
affiliations.parquet	5,271	2	Affiliation ID-name mapping
affiliation_geo.parquet	5,352	6	Affiliation, city, and country mappings
journals.parquet	46,237	2	Journal ID-name mapping
fields.parquet	21	3	Field ID-name mapping
intents.parquet	31	2	Citation intent ID-name mapping
cities.parquet	1,899	2	City ID-name mapping
countries.parquet	108	2	Country ID-name mapping

---

Citation Events

The main citation event files contain paper-level citation relations and citation semantics.

Important fields include:

Field	Description
citation_event_id	Unique citation event identifier
citing_paper_id	Identifier of the citing paper
cited_seed_paper_id	Identifier of the cited seed paper
citing_doi	DOI of the citing paper
citing_title	Title of the citing paper
citing_year	Publication year of the citing paper
citing_venue	Venue of the citing paper
primary_intent	Primary citation intent label
all_intents	List of citation intents associated with the citation event
contexts	Citation context text snippets
context_count	Number of citation contexts
intent_count	Number of intent annotations
is_influential	Whether the citation is marked as influential
has_semantic_evidence	Whether semantic evidence is available
field_id	Normalized field identifier

---

Seed Cited Papers

Seed papers represent highly cited papers used as cited targets.

Important fields include:

Field	Description
seed_paper_id	Unique seed paper identifier
doi	DOI of the seed paper
title	Title of the seed paper
publication_name	Publication venue
creator	Main creator or first author
citedby_count	Citation count
group	Broad research group
category	Research category
journal_clarivate	Clarivate journal name
is_seed_top5pct	Whether the seed paper belongs to the selected highly cited group

---

Knowledge Graph

The dataset includes a heterogeneous scholarly knowledge graph.

kg_nodes.parquet

Field	Description
node_id	Unique node identifier
label	Human-readable node label
node_type	Type of node
doi	DOI if the node corresponds to a paper
publication_name	Publication venue
group	Research group
citedby_count	Citation count
venue	Venue information
year	Publication year
primary_intent	Citation intent for citation-event nodes

kg_edges.parquet

Field	Description
source	Source node ID
target	Target node ID
edge_type	Type of graph relation

---

Intended Uses

This dataset can be used for:

1. Citation Intent Classification
2. Future Citation Prediction
3. AI Co-Scientist Hypothesis Generation
4. Scholarly Knowledge Graph Completion
5. Scientific Opportunity Discovery
6. Citation-based Scientific Recommendation
7. Evidence-grounded LLM Reasoning for Scientific Discovery

---

Dataset Construction

The dataset was constructed by integrating:

1. Citation event records
2. Citation contexts and citation intent labels
3. Highly cited seed paper metadata
4. Citing paper metadata
5. Journal, author, affiliation, city, country, and field normalization tables
6. Heterogeneous scholarly knowledge graph nodes and edges

The normalized files provide ID-based mappings for relational and graph-based experiments.

---

Dataset Structure

CitationHub/
├── affiliation_geo.parquet
├── affiliations.parquet
├── authors.parquet
├── citation_events.parquet
├── citation_events_enriched.parquet
├── citation_events_normalized.parquet
├── cities.parquet
├── citing_papers.parquet
├── citing_papers_normalized.parquet
├── countries.parquet
├── fields.parquet
├── intents.parquet
├── journals.parquet
├── kg_edges.parquet
├── kg_nodes.parquet
├── seed_cited_papers.parquet
└── seed_cited_papers_normalized.parquet

Ethical Considerations

This dataset is intended for scholarly analysis and research support. Citation-based metrics should not be used as the sole basis for evaluating researchers, institutions, journals, or countries.

---

Citation

@dataset{citationhub2026,
  title={CitationHub: An Intent-aware Scholarly Citation Knowledge Graph Dataset},
  author={Nam, Seohyun},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/Daniel0315/CitationHub}
}

---

License

CC BY 4.0

---