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	No
venue	Collaboration	64,482,050	64,405,068	64,698,700	27	year	No
publication	Collaboration	144,435,257	375,925,058	378,459,524	27	year	No
coauthor	Collaboration	31,547,264	656,908,420	3,032,478,219	27	year	No
reference	Collaboration	71,129,632	1,138,706,757	1,138,706,757	27	year	No
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
reddit	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	No
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.