Hugging Face's logo

Datasets:
brewcoua
/
genai-gdelt-framing

Tasks:
Text Classification
Modalities:
Tabular
Text
Formats:
parquet
Languages:
multilingual
Size:
1M - 10M
Tags:
generative-ai
ai-governance
news-framing
gdelt
computational-social-science
web-science
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
innovation_opportunity
float64
0.16
0.43
risk_safety
float64
0.16
1.63
regulation_governance
float64
0.52
2.13
rights_privacy
float64
0.04
0.16
economic_competition_labour
float64
0.14
0.54
misinformation_integrity
float64
0.03
0.28
month
int64
634
677
0.155273
0.160951
0.585732
0.036317
0.140769
0.146728
634
0.249195
0.186276
0.588074
0.057486
0.18319
0.051784
635
0.282407
0.25194
1.00089
0.063732
0.422593
0.056354
636
0.35288
0.319099
2.133522
0.081836
0.436808
0.084667
637
0.260969
0.375627
1.637417
0.105035
0.385478
0.043419
638
0.216288
0.410898
1.223768
0.123681
0.285301
0.039601
639
0.229095
0.351266
0.894647
0.068909
0.231268
0.049116
640
0.219999
0.314929
0.623412
0.064975
0.216125
0.032115
641
0.214739
0.432585
0.621184
0.075993
0.202776
0.039848
642
0.243689
0.44022
0.667672
0.075108
0.236458
0.029591
643
0.281234
0.409851
0.675071
0.093076
0.432179
0.053228
644
0.419091
1.631004
1.329761
0.111653
0.241108
0.118596
645
0.430259
1.044244
1.221072
0.127897
0.272591
0.059355
646
0.357984
0.850179
1.215662
0.118122
0.237726
0.069024
647
0.309351
1.519088
1.326662
0.160471
0.54262
0.079675
648
0.24111
0.679473
1.010594
0.08037
0.425898
0.125463
649
0.271023
0.654036
0.927668
0.09964
0.290319
0.132167
650
0.309911
0.804175
1.026445
0.0779
0.366723
0.076044
651
0.252007
0.701569
1.146335
0.067388
0.295904
0.070547
652
0.197979
0.508708
0.869769
0.072816
0.218547
0.216012
653
0.193636
0.384264
0.72098
0.052586
0.193998
0.280977
654
0.253046
0.66906
0.875065
0.105744
0.315187
0.080853
655
0.223841
0.637701
0.837827
0.075683
0.236971
0.095489
656
0.223
0.911712
0.781883
0.068936
0.210829
0.120458
657
0.256861
0.296165
1.199802
0.050556
0.205636
0.138376
658
0.223629
0.312297
0.799399
0.071022
0.256825
0.055037
659
0.250606
0.522907
1.517993
0.119338
0.259756
0.056452
660
0.232429
0.39846
1.132276
0.083362
0.295497
0.040901
661
0.22339
0.582271
1.251784
0.07103
0.192369
0.047204
662
0.213621
0.288235
0.719962
0.057853
0.232256
0.057595
663
0.194883
0.328748
0.68358
0.063222
0.206438
0.039263
664
0.270011
0.872643
1.03513
0.077464
0.251067
0.035486
665
0.203095
0.264673
0.834241
0.061289
0.176509
0.049024
666
0.192803
0.30286
0.668842
0.061491
0.313294
0.044916
667
0.169806
0.315873
0.610159
0.057919
0.197884
0.03873
668
0.175357
0.201766
0.570098
0.046694
0.153193
0.035274
669
0.218701
0.24224
0.704829
0.054578
0.230189
0.079461
670
0.180257
0.237566
0.524153
0.057904
0.173634
0.03383
671
0.210607
0.252032
0.575231
0.051
0.284157
0.052329
672
0.189857
0.239908
0.721869
0.047834
0.244048
0.081805
673
0.189982
0.516359
0.680976
0.064697
0.248868
0.0559
674
0.204005
0.37529
0.660947
0.073999
0.266722
0.060759
675
0.223462
0.306766
0.679029
0.050486
0.338045
0.055245
676
0.262718
0.246905
0.718424
0.072631
0.174855
0.045261
677

GenAI Governance Framing in Online News (GDELT 2.0, 2022–2026)

This dataset accompanies the paper:

Framing Generative AI Governance in Online News: A Longitudinal Analysis of 1.1 Million Articles (2022–2026)
Brewen Couaran, Yuvraj Singh Pathania, Arjun Rajesh Nair · 2026
PDF: paper/paper.pdf · Code: github.com/brewcoua/GenAI-GDELT
Companion site: brewcoua.github.io/GenAI-GDELT

Dataset overview

1,116,091 online news articles from the GDELT 2.0 Global Knowledge Graph, queried via Google BigQuery, covering November 2022 – June 2026 (44 months). Articles were selected using a two-condition filter: a 34-term generative AI lexicon AND a governance signal (policy keywords, GDELT thematic codes, or URL slug patterns).

Each article is annotated with a six-category governance frame taxonomy via a two-stage procedure:

  1. Keyword matching — multilingual dictionaries across 9 languages (318 English terms, 959 total)
  2. LaBSE embedding confirmation — cosine similarity between the article embedding and the frame's positive/negative pole centroids (FrameAxis method, Kwak et al. 2021)

A frame is confirmed when both the keyword flag fires and the embedding score is positive.

Configurations

articles — article-level records (1,116,091 rows)

Column Type Description
document_id string Source URL (GDELT DocumentIdentifier)
month string Publication month (YYYY-MM)
region string Source geography: US / EU / UK / Other
dominant_frame string Frame with highest normalized keyword count (null if unconfirmed)
kw_innovation_opportunity int8 1 = keyword match fired for this frame
kw_risk_safety int8
kw_regulation_governance int8
kw_rights_privacy int8
kw_economic_competition_labour int8
kw_misinformation_integrity int8
emb_innovation_opportunity float32 LaBSE bipolar embedding score (−1 to +1)
emb_risk_safety float32
emb_regulation_governance float32
emb_rights_privacy float32
emb_economic_competition_labour float32
emb_misinformation_integrity float32

A frame is confirmed when kw_* == 1 AND emb_* > 0. 40.8% of articles (455,349) are confirmed in at least one frame.

Note on the Regulation & Governance frame: Its base rate (16.6%) is likely structurally elevated because the corpus governance filter shares vocabulary with this frame's keyword dictionary. Cross-frame comparisons measure relative emphasis rather than absolute prevalence.

event_studies — milestone event study results (22 rows)

Pre/post window mean frame prevalence for 11 AI governance milestones (8 with windows ≥ 21 days + 3 short-window indicative events). Window length = min(d_prev, d_next, 90 days).

Column Description
milestone Milestone identifier slug
milestone_date Date (YYYY-MM-DD)
window_days Symmetric window length in days
side pre or post
n_articles Article count in this window-side
reg_governance … misinformation_integrity Mean confirmed frame prevalence

aggregates — summary statistics

Seven splits of precomputed aggregates: monthly_volume, monthly_frames, regional_frames, regional_frames_quarterly, tone_monthly, tone_by_frame, tone_by_region.

Frame taxonomy

Frame Description
Innovation & Opportunity Benefits, transformative potential, market opportunities
Risk & Safety Harms, threats, safety risks, existential concerns
Regulation & Governance Laws, oversight, compliance, institutional steering
Rights & Privacy Data protection, civil liberties, copyright, fairness
Economic Competition & Labour Jobs, automation, market dynamics, national AI competition
Misinformation & Integrity Deepfakes, disinformation, election integrity

Data provenance & license

Source data: GDELT 2.0 Global Knowledge Graph (Leetaru & Schrodt, 2013), released under Creative Commons Attribution (CC-BY). This derived dataset is released under CC-BY 4.0.

The raw article text is not included. document_id contains the source URL; full text must be retrieved independently. GDELT AllNames (named entities) and Quotations (direct speech) fields were used for frame assignment but are not redistributed here.

Citation 

@inproceedings{couaran_framing_2026,
  title     = {Framing Generative {AI} Governance in Online News:
               A Longitudinal Analysis of 1.1 Million Articles (2022--2026)},
  author    = {Couaran, Brewen and Pathania, Yuvraj Singh and Nair, Arjun Rajesh},
  year      = {2026},
  url       = {https://github.com/brewcoua/GenAI-GDELT},
}

Related links

Downloads last month
36
Number of rows:
1,116,527
Total file size:
90.5 MB