license: mit
tags:
- graph
- dynamic-graph
- temporal-graph
The GraphTide Real-World Dynamic Graph Datasets
GraphTide is a collection of large-scale, real-world dynamic graph datasets with authentic temporal information. It supports experimental evaluation of streaming graph algorithms, dynamic graph systems, and temporal graph analysis.
Source code, I/O and validation tools, and documentation are available in the GraphTide GitHub repository.
Graph Types
- Signal graphs record timestamped interactions, such as transactions, comments, commits, and trips.
- Update graphs record explicit edge insertions and deletions, allowing the graph state to be reconstructed over time.
Files are organized as:
signal_graphs/ Timestamped edge-event streams
update_graphs/ Explicit edge insertion/deletion streams
Datasets
All currently released graphs are unweighted. Unique Edges denotes
the number of distinct endpoint pairs, while Events denotes the
number of temporal signals or updates.
Signal Graphs
| Dataset | Domain | Vertices | Unique Edges | Events | Timestamps | Unit | Directed |
|---|---|---|---|---|---|---|---|
| ia-online-ads-criteo | Advertisement | 15,336,555 | 15,170,743 | 15,995,634 | 6,456,934 | seconds | No |
| soc-bitcoin | Blockchain | 24,575,382 | 86,063,840 | 122,948,162 | 19,444,110 | seconds | No |
| btc_stream_unweighted | Blockchain | 30,532,588 | 114,720,659 | 179,563,152 | 358,055 | seconds | No |
| ethereum | Blockchain | 78,945,641 | 236,100,055 | 781,489,512 | 10,200,000 | seconds | Yes |
| organization | Collaboration | 23,309,109 | 38,081,253 | 192,051,799 | 27 | year | Yes |
| venue | Collaboration | 64,482,050 | 64,405,068 | 64,698,700 | 27 | year | Yes |
| publication | Collaboration | 144,435,257 | 375,925,058 | 378,459,524 | 27 | year | Yes |
| coauthor | Collaboration | 31,547,264 | 656,908,420 | 3,032,478,219 | 27 | year | Yes |
| reference | Collaboration | 71,129,632 | 1,138,706,757 | 1,138,706,757 | 27 | year | Yes |
| semantic-scholar | Collaboration | 125,738,070 | 2,587,951,493 | 2,587,951,493 | 229 | year | Yes |
| rec-amazon-ratings | Recommendation | 2,146,057 | 5,743,258 | 5,838,041 | 3,329 | seconds | No |
| rec-amz-Sports-and-Outdoors | Recommendation | 2,469,419 | 3,268,695 | 3,268,695 | 4,786 | seconds | No |
| rec-amz-Cell-Phones-and-Acc | Recommendation | 2,580,723 | 3,447,249 | 3,447,249 | 4,724 | seconds | No |
| rec-amz-CDs-and-Vinyl | Recommendation | 2,064,957 | 3,749,004 | 3,749,004 | 6,041 | seconds | No |
| rec-amz-Home-and-Kitchen | Recommendation | 2,921,853 | 4,253,926 | 4,253,926 | 5,202 | seconds | No |
| rec-amz-Movies-and-TV | Recommendation | 2,289,561 | 4,607,047 | 4,607,047 | 6,004 | seconds | No |
| rec-amz-Clothing-Shoes-Jewelry | Recommendation | 4,253,272 | 5,748,920 | 5,748,920 | 4,209 | seconds | No |
| rec-amz-Electronics | Recommendation | 4,677,698 | 7,824,482 | 7,824,482 | 5,489 | seconds | No |
| rec-amz-Books | Recommendation | 10,356,390 | 22,507,155 | 22,507,155 | 6,296 | seconds | No |
| copresence-Thiers13 | Social | 328 | 43,496 | 18,613,039 | 8,938 | seconds | No |
| soc-youtube-growth | Social | 3,223,589 | 9,376,594 | 12,223,774 | 203 | seconds | No |
| sx-stackoverflow | Social | 2,584,164 | 28,183,518 | 47,903,266 | 41,484,769 | seconds | No |
| soc-flickr-growth | Social | 2,302,925 | 33,140,017 | 33,140,017 | 134 | seconds | No |
| bluesky | Social | 1,873,159 | 52,417,090 | 250,624,500 | 528,924 | seconds | No |
| Social | 74,000,011 | 746,954,361 | 8,989,320,299 | 168,851,154 | seconds | No | |
| llvm | Software | 243,737 | 1,019,582 | 2,453,587 | 526,032 | seconds | No |
| linux | Software | 193,269 | 1,078,218 | 2,838,279 | 1,192,006 | seconds | No |
| nyc_taxi_yellow | Transportation | 265 | 63,692 | 1,371,194,911 | 346,991,031 | seconds | Yes |
| flight_opensky | Transportation | 18,257 | 2,380,394 | 62,582,273 | 45,835,674 | seconds | Yes |
An additional ia-online-ads-criteo-attrs.bin variant is also
available in signal_graphs/.
Update Graphs
| Dataset | Domain | Vertices | Unique Edges | Events | Timestamps | Unit | Directed |
|---|---|---|---|---|---|---|---|
| wikipedia | Web | 62,489,082 | 401,854,218 | 1,294,550,620 | 183,259,984 | seconds | Yes |
| central-america | Road | 4,454,638 | 6,468,206 | 10,661,854 | 1,354,927 | seconds | Yes |
| australia-oceania | Road | 8,232,884 | 13,204,266 | 20,619,707 | 2,577,810 | seconds | Yes |
| south-america | Road | 26,944,705 | 43,659,791 | 71,807,857 | 6,686,033 | seconds | Yes |
| africa | Road | 43,817,014 | 65,355,044 | 95,425,505 | 8,085,311 | seconds | Yes |
| north-america | Road | 112,771,288 | 170,000,577 | 258,236,950 | 29,048,546 | seconds | Yes |
| asia | Road | 124,486,147 | 187,177,810 | 280,092,215 | 24,095,945 | seconds | Yes |
| europe | Road | 169,942,207 | 267,341,689 | 441,068,613 | 57,207,306 | seconds | Yes |
| planet | Road | 485,492,757 | 745,319,813 | 1,165,712,593 | 117,089,898 | seconds | Yes |
Download
Install the Hugging Face tools:
pip install -U huggingface_hub hf_xet
Download one dataset:
hf download DynamicGraphsProvider/GraphTide \
signal_graphs/soc-youtube-growth.bin \
--repo-type dataset \
--local-dir GraphTide-data
Download the complete collection:
hf download DynamicGraphsProvider/GraphTide \
--repo-type dataset \
--local-dir GraphTide-data
The complete collection is large, so downloading individual files is recommended.
Reading and Validation
Install the GraphTide tools:
git clone https://github.com/wheatman/GraphTide.git GraphTide-code
python -m pip install -e ./GraphTide-code
Read a header or load a graph:
from pybinding import read_header, read_bin
path = "GraphTide-data/signal_graphs/soc-youtube-growth.bin"
header = read_header(path)
header, data = read_bin(path)
Large datasets should be processed with the streaming C++ I/O interface rather than loaded completely into memory.
Validate a binary graph:
cd GraphTide-code
python -m pybinding.validate /path/to/dataset.bin
For an update graph, the optional stateful check also validates the insertion/deletion sequence:
python -m pybinding.validate /path/to/dataset.bin \
--check-update-sequence
Binary Format
Each file contains a 32-byte little-endian header followed by tightly packed records:
src, dst, [timestamp], [weight], [info]
The header specifies the graph type, direction, data types, number of
vertices, unique edges, and temporal records. For update graphs,
info=0 denotes insertion and info=1 denotes deletion.
See the complete binary format specification.
Contributing a Dataset
Providing the original parser is encouraged but not required.
Before uploading a large file, open a New dataset proposal describing the source, license, graph semantics, timestamps, cleaning decisions, and approximate size.
After approval, validate the graph and upload it through a Hugging Face pull request:
hf auth login
hf upload DynamicGraphsProvider/GraphTide \
/path/to/dataset.bin signal_graphs/DATASET_ID.bin \
--repo-type dataset \
--create-pr \
--commit-message "Add DATASET_ID dynamic graph"
Use update_graphs/DATASET_ID.bin for an update graph, and link the
Hugging Face pull request from the original GitHub proposal.
See the complete contribution guidelines.
For questions, use the GraphTide issue tracker.