CitationHub is an interactive exploration system built on IDCite, a large-scale multidisciplinary citation-event dataset. It lets readers search influential seed papers and authors, inspect citation intents and contexts, and explore geographic, network, and knowledge-graph views of the same underlying resource.
Website URL: https://citation-hub-website.vercel.app
This repository explains how CitationHub is organized and how it uses IDCite. It is written for readers who want to understand the system, not to rebuild or redeploy it.
🎥 Watch the CitationHub System in action: Watch the interactive demo
The home page is the entry point to the system.
- The top navigation lists eight modules: Home, Search, Authors, Analytics, Geographic Map, Citation Network, Knowledge Graph, and SPARQL.
- The hero search box switches between Papers (title or DOI) and Authors. Under the paper search box, Advanced search opens a fielded form (title, author, journal, publication year).
- Dataset Overview reports the scale currently served by the interactive system (seed papers, citation events, citing papers, authors, journals, countries, fields, and citation intents).
- Explore links to the six analysis modules.
The numbers on the home page describe the live index. The full IDCite release is larger and is documented in Data layer.
Each navigation tab is a separate view over the same IDCite tables.
The landing page introduces the corpus, provides paper/author search, and summarizes dataset scale. The Explore cards jump to Search, Authors, Analytics, Geographic Map, Knowledge Graph, and SPARQL. The paper search box includes an Advanced search link to /search/advanced.
Keyword search over seed-paper titles and DOIs, with Field, Country, and Journal filters, pagination, and CSV export. Opening a result leads to the paper detail page. The same page also accepts the fielded query produced by Advanced search (title, author, journal, year_min, year_max).
/search/advanced is a fielded form over the seed-paper catalogue, reached from Home (and from the paper search bar). It is not a separate data source: the same /search results page and /api/search filters serve both the simple box and this form.
- Keywords match tokens in the title or DOI.
- Title, Author(s), and In journal(s) each search that bibliographic field.
- Filled fields are combined with AND. Within a field, every whitespace-separated word must appear (substring match). Example:
deep learningkeeps papers whose title or DOI contains both words. - Author strings in IDCite follow Scopus style (
Studenski S.), not a full given name. - Date published is the seed paper’s cover year, not the year it was cited. Presets: Any date, Since 2020, 2015–2019, 2010–2014, Before 2010, or a custom YYYY–YYYY range.
Author-name search grouped by author identifier. Each card shows affiliation, country, primary field, seed-paper count, and total citations. Opening a card leads to the author detail page: profile metadata, seed papers, citation-intent distribution, citations over time, and top citing venues.
Global statistics across citation events:
- citation-intent distribution
- share of citations marked influential
- intent trends by year
- top citing venues
A world map of seed-paper counts with three switchable layers — by country (choropleth with country names), by city, and by affiliation (bubbles sized by paper count). All three layers are labelled automatically. Ranked lists of top countries, cities, and affiliations sit under the map. IDCite names cities and affiliations but carries no coordinates; CitationHub places them with a GeoNames lookup shipped with the application.
A force-directed network for a selected seed paper. Citing papers are nodes colored by primary citation intent (background, uses, similarities, motivation, differences, future_work, extends). Drag to pan, scroll to zoom, and click a node for details.
A heterogeneous graph around a seed paper: papers, citation events, authors, journals, affiliations, cities, countries, fields, and intents. A second tab, Citation Event Schema, shows the class/relation diagram used by the graph and SPARQL views.
A read-only SPARQL explorer over the RDF knowledge graph (about 25 million triples). Users can run prepared example queries or write SELECT / ASK / CONSTRUCT / DESCRIBE queries, inspect an ontology panel, and download results as CSV. Write operations are not allowed.
Not a navigation tab, but reachable from Search, Citation Network, and Knowledge Graph. It shows bibliographic metadata, intent distribution, citing papers, co-cited papers, citation-context snippets, a knowledge-graph shortcut, and CSV export of that paper’s citation events.
More screen-level detail is in docs/features.md.
CitationHub is the application layer. IDCite is the data layer: 17 Apache Parquet tables of citation events, publication metadata, normalized scholarly entities, and a supplementary knowledge graph.
| Measure (IDCite release) | Value |
|---|---|
| Citation events | 1,857,503 |
| Citing papers | 1,467,045 |
| Seed (cited) papers | 23,479 |
| ESI fields | 21 |
| Q1 journals in the sampling frame | 105 |
| Knowledge-graph nodes / edges | 3,418,433 / 6,855,117 |
| Data snapshot | November 2025 |
The interactive system reads a working subset of those tables (normalized seed papers, normalized citation events, the author lookup, and the knowledge-graph node/edge files). Geographic Map uses city, country, and affiliation names already on the seed-paper table, plus a separate coordinate lookup derived from GeoNames — not from IDCite. The remaining tables are part of the public dataset release.
Where to get the 17 Parquet files
| Resource | Location |
|---|---|
| IDCite Version 3 (Zenodo) | https://doi.org/10.5281/zenodo.20796923 |
| Processed dataset mirror (Hugging Face) | https://huggingface.co/datasets/Daniel0315/IDCite |
| Dataset Construction pipeline | https://github.com/SeohyunNam/MDCite |
File-by-file inventory and schemas: docs/data.md.
CitationHub separates the public interface from the data and query services.
Browser
│
▼
Next.js frontend (search, analytics, maps, graphs, SPARQL editor)
│ HTTPS API
▼
FastAPI backend
├── Parquet tables (pandas / DuckDB)
└── read-only SPARQL proxy
│
▼
SPARQL engine (QLever) serving the RDF knowledge graph
- The frontend is a Next.js application: paper/author search, charts, a world map (country choropleth plus city and affiliation bubbles), force graphs, and the SPARQL editor.
- The backend loads IDCite Parquet files and answers overview, search, paper, author, analytics, geo, and knowledge-graph requests.
- The RDF layer is produced from
kg_nodes.parquetandkg_edges.parquet. QLever indexes those triples; the backend forwards read-only SPARQL queries and rejects update operations.
This repository does not include deployment hostnames, internal paths, or credentials. Logical design is in docs/architecture.md.
The SPARQL page queries a scholarly RDF graph with namespaces:
- resources:
http://citationhub.org/id/ - ontology:
http://citationhub.org/ontology/(ch:)
Classes: SeedPaper, CitingPaper, CitationEvent, Author, Journal, Affiliation, City, Country, Field, Intent
Relations: hasAuthor, hasAffiliation, belongsToField, publishedIn, publishedInVenue, locatedInCity, locatedInCountry, hasCitedPaper, hasCitingPaper, hasPrimaryIntent
The SPARQL engine is QLever. Example queries: examples/sparql/. Ontology notes and references: docs/sparql.md.
- CitationHub system: https://citation-hub-website.vercel.app
- IDCite construction pipeline: https://github.com/SeohyunNam/MDCite
- IDCite dataset (Zenodo v3): https://doi.org/10.5281/zenodo.20796923
- IDCite mirror (Hugging Face): https://huggingface.co/datasets/Daniel0315/IDCite
- IDCite project page: https://seohyunnam.github.io/IDCite-Website/
Seohyun Nam, Department of Computer Engineering, Chung-Ang University
Seohyun0315@cau.ac.kr
Documentation in this repository is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0). Reuse of IDCite itself follows the dataset license and the terms of the upstream scholarly sources.