Datasets:

andreas-helgesson
/

gdelt-big

Modalities:

Text

Formats:

text

Size:

100M - 1B

Libraries:

Datasets

License:

Dataset card Data Studio Files Files and versions Community

Dataset Viewer (First 5GB)

Auto-converted to Parquet

Split (3)

train · ~389M rows (showing the first 283M)

text stringlengths 11 16
0 0 251 0 0
1 1 252 0 0
2 1 251 0 0
3 1 251 0 0
4 1 251 0 0
5 1 251 0 0
6 1 251 0 0
1 1 251 0 0
6 0 253 0 0
6 0 254 0 0
7 2 255 0 0
8 0 256 0 0
6 1 252 0 0
9 0 256 0 0
10 0 256 0 0
4 0 256 0 0
0 0 256 0 0
11 0 256 0 0
6 0 256 0 0
1 0 256 0 0
12 0 256 0 0
13 0 256 0 0
14 0 256 0 0
15 0 256 0 0
6 0 254 0 0
2 0 256 0 0
5 1 252 0 0
4 1 252 0 0
3 1 252 0 0
6 1 257 0 0
0 1 257 0 0
2 0 256 0 0
16 0 256 0 0
11 0 256 0 0
6 0 256 0 0
1 0 256 0 0
15 0 256 0 0
17 0 258 0 0
7 0 258 0 0
6 0 258 0 0
18 0 258 0 0
2 1 259 0 0
3 1 259 0 0
4 1 259 0 0
5 1 259 0 0
6 1 259 0 0
1 1 259 0 0
2 1 260 0 0
3 1 260 0 0
4 1 260 0 0
5 1 260 0 0
6 1 260 0 0
1 1 260 0 0
2 1 252 0 0
19 0 261 0 0
0 1 262 0 0
4 0 261 0 0
19 0 263 0 0
1 0 264 0 0
20 0 264 0 0
6 0 265 0 0
10 2 253 0 0
5 2 253 0 0
21 2 253 0 0
22 2 253 0 0
6 2 253 0 0
1 2 253 0 0
11 2 260 0 0
6 2 266 0 0
6 0 264 0 0
23 0 267 0 0
24 0 267 0 0
4 0 267 0 0
23 0 268 0 0
6 0 268 0 0
24 0 268 0 0
4 0 268 0 0
23 0 259 0 0
6 0 259 0 0
24 0 259 0 0
4 0 259 0 0
6 1 254 0 0
6 0 267 0 0
5 0 264 0 0
25 0 264 0 0
20 0 253 0 0
4 0 263 0 0
0 0 263 0 0
2 0 269 0 0
16 0 269 0 0
26 0 269 0 0
6 0 269 0 0
1 0 269 0 0
20 0 269 0 0
6 3 270 0 0
6 0 254 0 0
6 0 271 0 0
6 2 272 0 0
6 0 273 0 0
26 0 273 0 0

End of preview. Expand in Data Studio

GDELT-big

Description

GDELT-big is a Temporal Knowledge Graph (TKG) dataset following a format compatible with FinDKG and RE-GCN.

Compared to the datasets used by FinDKG (FinDKG, FinDKG-full) and RE-GCN (ICEWS18, ICEWS14, ICEWS05-15, GDELT), GDELT-big offers superior graph density (in regards to the number edges per node and timestep) and covers a much longer time span. See below table for key characteristics:

	ICEWS18	ICEWS14s	ICEWS05-15	GDELT	FinDKG	FinDKG-full	GDELT-big
Entities	23033	7128	10488	7691	13645	13645	1302
Edge Types	256	230	251	240	15	15	5
No. Edges (Train)	373018	74845	368868	1734399	119549	222732	388876937
No. Edges (Val)	45995	8514	46302	238765	11444	9404	48609616
No. Edges (Test)	49545	7371	46159	305241	13069	10013	48609617
No. Timesteps (Train)	240	304	3243	2138	100	234	483
No. Timesteps (Val)	30	30	404	265	13	13	78
No. Timesteps (Test)	34	31	370	348	13	14	68
No. Edges/Timestep	1541	249	115	828	1143	928	772808
Timestep Resolution	24 hours	24 hours	24 hours	15 min	1 week	1 week	1 week

GDELT-big is assembled from the GDELT 1.0 Global Knowledge Graph (GKG) data stream, and spans the period 2013-04-01 to 2025-04-14, with entity relations aggregated per week (time2id refers to the Monday of each such week). It is chronologically split into training, validation, and testing, similarly to the above mentioned datasets. Its entities are extracted from a pre-defined list of nations and Fortune 500 companies.

GDELT-big was created primarily due to the relative sparsity of existing TKG datasets. Of the other datasets listed in the above table, on average, less than 7% of all nodes have a degree > 0 in any randomly selected timestep. For GDELT-big, this value is 22%. As such, the nodes in GDELT-big are far more interconnected than in previously available datasets.

Key Features

Extensive Coverage: The dataset spans 12 years, offering broad insights into how the news landscape for the specific entities evolve over time.
Superior Density: The TKG has far more edges per node in each timestep, enabling new insights to be drawn from TKG predictions.
Time-Series Analysis Ready: The dataset is provided in a format compatible with previous research in the area.

Citation

If you use the GDELT-big dataset in your research, please cite our work as follows:

http://hdl.handle.net/20.500.12380/309518

@blomhelgesson{BlomHelgesson2025,
  author    = {Blom, Axel and Helgesson, Andreas},
  title     = {Financial News Event Prediction Using Temporal Knowledge Graphs},
  school    = {Chalmers University of Technology},
  year      = {2025},
  month     = {June},
  type      = {Master's Thesis in Data Science \& AI},
  note      = {Department of Mathematical Sciences}
}