OpenAlex Dataset

A complete mirror of the OpenAlex scholarly knowledge graph, plus a pre-built graph index and tile repository for client-side visualization.

Subsets

Snapshot Records (full OpenAlex data)

Config	Split	Records	Description
complete	records	~583M	All entity types combined
works (default)	records	~449M	Scholarly works — papers, articles, books, datasets
authors	records	~108M	Author profiles with affiliations and metrics
institutions	records	~103K	Universities, research institutes, hospitals
sources	records	~255K	Journals, repositories, conferences
topics	records	~4.5K	Research topics (auto-classified)
subfields	records	252	Academic subfields
fields	records	26	Academic fields
domains	records	4	Top-level domains
concepts	records	~65K	Subject concepts (legacy, replaced by topics)
funders	records	~32K	Funding organizations
publishers	records	~11K	Publishing companies

Each record is the full OpenAlex JSON matching the REST API schema. Fields vary by entity type — see OpenAlex docs for details.

Source files: gzip-compressed JSON Lines from the OpenAlex S3 snapshot (January 2026), date-partitioned under snapshot/<entity>/YYYY-MM-DD/.

Graph Index (compact IDs)

Config	Split	Records	Schema	Description
index	works	~449M	{id, n}	Compact integer ID + title
	authors	~108M	{id, n}	Compact integer ID + name
	institutions	~103K	{id, n}	Compact integer ID + name
	sources	~255K	{id, n}	Compact integer ID + name
	topics	~4.5K	{id, n}	Compact integer ID + name
	subfields	252	{id, n}	Compact integer ID + name
	fields	26	{id, n}	Compact integer ID + name
	domains	4	{id, n}	Compact integer ID + name
	concepts	~65K	{id, n}	Compact integer ID + name
	funders	~32K	{id, n}	Compact integer ID + name
	publishers	~11K	{id, n}	Compact integer ID + name

Pipeline-generated compact IDs (0-indexed integers) mapped from OpenAlex IDs. Used internally by the graph visualization and tile system.

Utility Configs

Config	Split	Schema	Description
search	records	{w, ids}	Inverted word index — maps lowercased tokens to compact entity IDs
names	records	{id, name}	Display names keyed by compact ID, sharded for tile serving

Quick Start

from datasets import load_dataset

# Load full OpenAlex works records
works = load_dataset("Mearman/OpenAlex", "works", split="records")

# Stream the complete dataset (all entity types, ~434 GB compressed)
ds = load_dataset("Mearman/OpenAlex", "complete", split="records", streaming=True)

# Load a single small entity type
topics = load_dataset("Mearman/OpenAlex", "topics", split="records")

# Load the compact graph index for authors
authors_index = load_dataset("Mearman/OpenAlex", "index", split="authors")

# Search index
search = load_dataset("Mearman/OpenAlex", "search", split="records")

Download specific directories

from huggingface_hub import snapshot_download

# Download just the raw snapshot for one entity type
snapshot_download(
    "Mearman/OpenAlex",
    repo_type="dataset",
    allow_patterns="snapshot/topics/**",
    local_dir="./openalex",
)

# Download tiles for 2D visualization
snapshot_download(
    "Mearman/OpenAlex",
    repo_type="dataset",
    allow_patterns=["tiles/manifest.json", "tiles/nodes/*", "tiles/coords/2d/*", "tiles/spatial/2d/*"],
    local_dir="./openalex",
)

About OpenAlex

OpenAlex is a free, open catalogue of the global research system — scholarly works, authors, institutions, journals, and the connections between them. It is maintained by OurResearch and available under a CC0 public domain license.

The snapshot/ directory in this repository is a mirror of the OpenAlex S3 snapshot (AWS Open Data Registry). The upstream snapshot is updated approximately monthly.

Citation

If you use this data, please cite OpenAlex:

Priem, J., Piwowar, H., & Orr, R. (2022). OpenAlex: A fully-open index of scholarly works, authors, venues, institutions, and concepts. ArXiv. https://arxiv.org/abs/2205.01833

See also: OpenAlex documentation · API · GitHub

Repository Structure

snapshot/               # Raw OpenAlex snapshot (gzipped JSONL, date-partitioned)
├── works/              # ~449M scholarly works (~750 GB compressed)
│   └── YYYY-MM-DD/part_NNNN.jsonl.gz
├── authors/            # ~108M author profiles (~64 GB compressed)
├── institutions/       # 103K institutions
├── sources/            # 255K journals/repositories
├── concepts/           # 65K subject concepts
├── topics/             # 4.5K topics
├── funders/            # 32K funders
├── publishers/         # 11K publishers
├── domains/            # 4 top-level domains
├── fields/             # 26 fields
└── subfields/          # 252 subfields

index/
├── entities/          # Small entity types (JSONL.gz: {id, n})
├── mappings/          # Binary ID mappings (OpenAlex ID → compact ID)
├── authors/           # Author index (104 shards, JSONL.gz)
├── works/             # Work index (1,098 shards, JSONL.gz)
├── edges/             # Citation + collaboration edges (adjacency lists)
├── adj/               # Compressed adjacency shards
└── search/            # Inverted word index for entity search

layout/
├── coords-2d.bin      # 2D force-directed layout coordinates
├── coords-3d.bin      # 3D force-directed layout coordinates
├── layout-2d.json     # Layout metadata
├── layout-3d.json
└── density-2d.png     # 4K density visualization

tiles/
├── manifest.json
├── lod/{dim}d/        # Level-of-detail with embedded coordinates
├── nodes/             # Entity metadata by compact ID range
├── coords/{dim}d/     # Position data by compact ID range
├── spatial/{dim}d/    # Spatial index (Z-order curve tiles)
├── edges/             # Edge tiles by source compact ID range
├── lookup/            # Compact ID ↔ OpenAlex ID mapping
├── names/             # Display name shards
├── search/            # Symlinked from index/search
└── adj/               # Symlinked from index/adj

Pipeline

Built using the OpenAlex Pipeline on Supercomputing Wales (Falcon cluster, 192-core AMD Genoa nodes) as part of PhD research at Bangor University:

1. Download — S3 snapshot sync via rclone
2. Extract — isal-accelerated gzip decompression
3. Index — Parallel entity indexing, ID mapping, edge extraction, search index
4. Layout — Multi-level coarsening + Barnes-Hut force-directed simulation (2D & 3D)
5. Tiles — Dimension-agnostic spatial tiling for progressive client-side loading

License

The underlying OpenAlex data is CC0 (public domain). This processed derivative is released under the same license.

Status

🚧 In progress — Raw snapshot and index uploaded. Layout and tiles will be added as pipeline jobs complete.