Cool graph datasets (beyond Neo4j’s official examples)
The Neo4j getting-started example data page is already a strong mix of story graphs (Movies, Game of Thrones), recommendations, retail, investigations (FinCEN, Panama/Paradise), social, recipes, transit, and IT.
Here are other datasets in the same spirit — ready-to-play, narrative, or “follow the connections” graphs — plus the big public catalogs if you want scale.
More Neo4j-ready examples
These are already modeled for Neo4j (dumps, Cypher loaders, or Browser guides), so they drop in like the official samples:
- Twitch streamers — streamer–game–viewer graph (
neo4j-graph-examples/twitch) - ICIJ Offshore Leaks — companies, officers, addresses, jurisdictions; same family as Panama/Paradise
- Public explorer: https://offshoreleaks.icij.org/
- Repo:
neo4j-graph-examples/icij-offshoreleaks - OpenStreetMap / NYC Central Park — geo graph of ways, nodes, amenities (
neo4j-graph-examples/openstreetmap) - Entity resolution demo — record-linkage graph (
neo4j-graph-examples/entity-resolution) - William Lyon’s extra packs — Yelp reviews, TrumpWorld org chart, 2016 FEC campaign finance, IoT
- https://github.com/johnymontana/neo4j-datasets
- OpenSanctions + Offshore Leaks merge — sanctioned people linked into the ICIJ offshore graph (
opensanctions/offshore-graph) - Enron email corpus as a graph — ICIJ used this as a Datashare + Neo4j demo (~500k emails). Good for “who talked to whom” path queries
- NASA knowledge graph — older Browser guide (
:play nasa) - GraphGists — community models across industries: https://neo4j.com/graphgists
Quality varies, but lots of odd, specific graphs
Also still useful from the official page if you skipped them: WordNet, Neoflix, football transfers, UK companies + land + donations, FAERS healthcare, London Tube.
Pop culture & story graphs
Same energy as Movies / GoT:
- Star Wars character interactions (episodes 1–7) — scene co-occurrence
- https://github.com/evelinag/StarWars-social-network
- Also bundled in the
networkdataR package - Les Misérables co-appearance network — Knuth’s classic weighted character graph (easy Gephi/GML download)
- Shakespeare play networks — character speech/interaction graphs per play
- Marvel / comic character networks — widely recirculated; look under movie/comic network collections on Harvard Dataverse and
networkdata - Jazz musicians collaboration network
These are small, visual, and great for community detection, shortest paths, and “who bridges two factions.”
Investigations, money, and power
If FinCEN / Panama clicked:
- Full ICIJ Offshore Leaks Database — Panama, Paradise, Pandora, Bahamas, etc. Downloadable; ICIJ themselves run it on Neo4j
- OpenSanctions — PEPs, sanctions, crime lists as entities + relationships. Combine with offshore data
- US FEC campaign finance (2016 pack above) — donors → committees → candidates
- Political books co-purchase (
polbookson Netzschleuder) — Amazon “customers also bought” among political titles. Tiny and famous - Signed trust networks — Bitcoin OTC/Alpha, Slashdot friend/foe, Epinions (SNAP). Positive/negative edges are fun for balance theory
Places, systems, and “how the world is wired”
- OSMnx / global urban street networks — GraphML for thousands of cities (Harvard Dataverse). Same idea as the London Tube example, but worldwide
- OpenStreetMap extracts — any city as a routing / amenity graph
- Airport / air-travel reachability — SNAP “reachability” plus several Netzschleuder airport graphs
- US / European power grids — Network Repository and SNAP road/power graphs
- Yelp Open Dataset — businesses, users, reviews, friends (large; Lyon has a Neo4j cut)
- Last.fm / HetRec — users–artists–tags (Neo4j GDS client also ships a LastFM hetero graph)
- Product Space / Atlas of Economic Complexity — products linked by export similarity. Beautiful “what does this country make next?” graph
- Food webs — Florida Bay, New Zealand streams, etc. (Netzschleuder
fresh_webs, SNAP Florida Bay) - C. elegans connectome — 131 neurons, spatial coordinates. Smallest “real brain” people actually query
Classic network-science catalogs
When you want lots of graphs rather than one polished demo:
| Collection | What you get | Link |
|---|---|---|
| SNAP | Social, citations, co-authorship, web, Amazon co-purchase, roads, Wikipedia talk/votes, temporal, signed | https://snap.stanford.edu/data |
| Netzschleuder | ~286 curated datasets, many formats (GraphML/GML/CSV), good metadata | https://networks.skewed.de |
| Network Repository | Thousands of graphs + in-browser viz | https://networkrepository.com |
| KONECT | 1,000+ networks, stats and plots | http://konect.cc |
| ICON | Index of Complex Networks (discovery index) | University of Colorado / Clauset group |
| networkdata (R) | 987 datasets / 2,260 networks: animals, covert orgs, tennis, Star Wars, movies | https://schochastics.github.io/networkdata/ |
| Gephi sample pack | Les Mis, internet AS, Hypertext 2009 contacts, jazz, school friendships | https://docs.gephi.org/desktop/User_Manual/Datasets |
| Awesome Graph Datasets | Typed list: plain, attributed, bipartite, text-attributed, temporal | https://github.com/HKBU-LAGAS/Awesome-Graph-Datasets |
SNAP highlights that feel “cool” rather than just large: ego-Facebook circles, wiki-Vote, Higgs Twitter, Amazon co-purchase, ca-HepTh / CondMat collaborations, soc-Pokec, LiveJournal.
Knowledge graphs and ML-scale
- WordNet — already on the Neo4j demo server
- Wikidata / ogbl-wikikg2 and WikiKG90M — entity–relation–entity triples at serious scale (Open Graph Benchmark)
- ogbn-mag / MAG240M — papers, authors, institutions, fields (Microsoft Academic Graph slice)
- ogbl-biokg / BioSNAP — drugs, diseases, proteins, side effects
- Cora / CiteSeer / PubMed — citation + bag-of-words features; Neo4j GDS client can load Cora and Karate Club directly
- ogbn-products — Amazon co-purchase with category labels (~2.4M nodes)
- Wikipedia hyperlink graph — English wiki page–page links (SNAP
enwiki-2013or dedicated Wikipedia-as-Neo4j projects)
OGB portal: https://ogb.stanford.edu
Practical load path into Neo4j
- Prefer official dumps /
:guidedatasets when you want a model + queries already written. - For SNAP / Netzschleuder / Network Repository: download edge lists or GraphML →
LOAD CSVorapoc.import.graphml. - For OSM:
neo4j-graph-examples/openstreetmapor export via OSMnx to GraphML. - For ICIJ / OpenSanctions: start with their CSVs or the existing Neo4j example repos rather than scraping the live sites.
- Watch size. Stack Overflow full dump and WikiKG/MAG240M need real RAM; Movies/GoT/Karate/Les Mis/POLE do not.