ReGKG-2025: A GKG-Compatible Reconstruction of the GDELT Global Knowledge Graph for the Missing Period of 14 June to 2 July 2025
Organizations: HSE University, Moscow, Russia
Abstract
The June 2025 outage in the Global Database of Events, Language, and Tone (GDELT) collection pipeline left a gap in the Global Knowledge Graph (GKG), interrupting article-level news metadata used in downstream research. ReGKG-2025 is an archive-derived sample of 281,620 reconstructed article records covering all 1,665 missing 15-minute batches. Articles were recovered from Common Crawl and the Wayback Machine using source quotas estimated from surrounding native GKG data. Text was screened and normalized, non-English bodies were translated, and flagged records were manually reviewed. Eligible articles were converted to the 27-column GKG 2.1 layout. The versioned package includes record-level provenance, archive replay and WARC locators, source summaries, source-by-day calibration weights, and the complete manual decision journal. The sample reaches about 14% of modeled native volume and represents sources accounting for 84% of selected reference record mass. In an intact-week test, 41% of recovered articles matched native GKG records; field agreement on matched pairs is close to the processor baseline (tone MAE 1.69). The dataset supports retrospective content analysis and source-level aggregates; its tone and theme distributions differ from native GKG, and it does not support exhaustive counts. Data: https://doi.org/10.5281/zenodo.23964265.
Figures & tables
| Property | Value |
|---|---|
| Version | 1.0.0 |
| Article records | 281,620 |
| Represented sources | 4,301 |
| Occupied / target batches | 1,665 / 1,665 |
| Selected-source identities | 77.52% |
| Selected reference mass covered | 83.94% |
| Field group | Status | Construction |
|---|---|---|
| Record ID, date, source, URL | Constructed | Metadata and publication bucket |
| Persons, organizations, locations | Approximate | Entity recognition and geographic resolution |
| Themes | Approximate | Phrase and context rules |
| Tone | Approximate | Lexical scores and densities |
| GCAM | Approximate | Dictionary counts and weighted scores |
| Counts, date mentions, amounts, quotations, all names | Approximate | Extracted values and text spans |
| File | Contents and size |
|---|---|
| data.zip | GKG batch archives and master list |
| 1.10 GiB | |
| record_provenance.parquet | Record-level provenance |
| 22.61 MiB | |
| archive_provenance.parquet | Archive replay/range locators |
| 31.35 MiB |
| Comparison | Release | Random | Native |
|---|---|---|---|
| Source shares, all sources | 0.382 | 0.054 | 0.127 |
| Source shares, selected sources | 0.358 | 0.045 | 0.100 |
| Weekly profile (minute precision) | 0.171 | 0.033 | — |
| Weekly profile (all) | 0.244 | 0.030 | — |
| Source day | 0.503 | 0.206 | — |
| Source day, source totals matched | 0.292 | 0.144 | — |
| Comparison | Themes | Countries | Tone |
|---|---|---|---|
| Complete populations | |||
| Pre / gap | 0.275 | 0.126 | 0.105 |
| Post / gap | 0.278 | 0.149 | 0.108 |
| Pre / post | 0.075 | 0.078 | 0.012 |
| Common sources, gap source weights | |||
| Pre / gap | 0.268 | 0.126 | 0.103 |
| Population | Pairs | Tone MAE | Tone CCC | Persons | Organizations | Locations | Themes |
|---|---|---|---|---|---|---|---|
| Processor study: English | 193,550 | 1.790 | 0.749 | 0.646 | 0.417 | 0.654 | 0.617 |
| Intact week: all pairs | 5,918 | 1.690 | 0.806 | 0.673 | 0.456 | 0.648 | 0.638 |
| Intact week: English source | 5,891 | 1.689 | 0.807 | 0.674 | 0.458 | 0.650 | 0.639 |
| Intact week: balanced subset | 2,081 | 1.697 | 0.799 | 0.674 | 0.441 | 0.650 | 0.634 |
| Processor study: title/length | 54,485 | 1.470 | 0.830 | 0.748 | 0.483 | 0.727 | — |
| Intact week: title/length | 1,848 | 1.470 | 0.851 | 0.725 | 0.506 | 0.686 | 0.671 |
| Task | Suitability | Conditions |
|---|---|---|
| Retrospective content series | Supported with controls | Report sample size, source composition, timestamp precision, and regime boundaries. |
| Source-level daily comparison | Conditional | Restrict to supported cells; report weights and missing-source mass. |
| GKG-format pipeline testing | Supported | Read all 27 fields and distinguish four unavailable media columns from no text matches. |
| Gap-handling benchmarks | Conditional | Evaluate against observed intact periods; preserve archive selection and document-version uncertainty. |
| Exhaustive article/event counts | Unsupported | Modeled-volume calibration does not recover absent publications or identify inclusion probabilities. |
| Real-time trading or monitoring | Unverified | Publication and archive capture times do not establish historical first availability. |
Appendix figures & tables16 assets
Supplementary material from the paper’s appendix.
Appendix
| Outcome | Records |
|---|---|
| Automated Preparation | |
| Ready | 281,001 |
| Flagged for review | 1,545 |
| Rejected | 2,489 |
| Translation failed | 101 |
| Manual Review | |
| Comparison | Mean TV | 2.5th | 97.5th | |
|---|---|---|---|---|
| Independent Reference Draws | ||||
| All sources | 281,620 | 0.054 | 0.053 | 0.055 |
| Selected sources | 281,620 | 0.045 | 0.044 | 0.046 |
| Temporal, all | 281,620 | 0.030 | 0.029 | 0.031 |
| Temporal, minute precision | 237,465 | 0.033 | 0.031 | 0.034 |
| Source day | 281,620 | 0.206 | 0.205 | 0.207 |
| Method | Minute | Day only |
|---|---|---|
| JSON-LD | 172,942 | 2,162 |
| HTML metadata | 55,288 | 1,616 |
| Labeled time element | 4,452 | 501 |
| First time element | 4,783 | 177 |
| Extractor date fallback | 0 | 39,699 |
| Comparison | TV | JSD | Ideal TV |
|---|---|---|---|
| Source Shares | |||
| All native sources | 0.382 | 0.196 | 0.054 |
| Selected sources | 0.358 | 0.175 | 0.045 |
| Temporal Profile | |||
| All records | 0.244 | 0.091 | 0.030 |
| Minute precision | 0.171 | 0.035 | 0.033 |
| Source | Native (%) | Output (%) | (pp) | TV (pp) |
|---|---|---|---|---|
| iheart.com | 6.04 | 8.63 | +2.59 | 1.30 |
| yahoo.com | 3.28 | 4.03 | +0.75 | 0.38 |
| marketscreener.com | 0.43 | 0.00 | -0.43 | 0.22 |
| dailymail.co.uk | 0.39 | 0.00 | -0.39 | 0.20 |
| boredpanda.com | 0.41 | 0.06 | -0.35 | 0.18 |
| tvguide.co.uk | 1.56 | 1.22 | -0.34 | 0.17 |
| Rank | Sources | Present (%) | Native (%) | Output (%) | Median (%) |
|---|---|---|---|---|---|
| 1–10 | 10 | 80.00 | 14.99 | 16.45 | 7.90 |
| 11–100 | 90 | 65.56 | 13.05 | 8.67 | 3.22 |
| 101–1,000 | 900 | 88.00 | 33.46 | 40.04 | 23.65 |
| 1,001+ | 11,459 | 30.04 | 38.49 | 34.84 | 0.00 |
| Population | Publication matches | Coverage (%) | Normalized matches |
|---|---|---|---|
| Full pool | 3,432 | 0.731 | 6,058 |
| Balanced subset | 1,314 | 0.280 | 2,101 |
| Reference rank | Sources | Native URLs | Matches | Coverage (%) |
|---|---|---|---|---|
| 1–10 | 10 | 127,612 | 369 | 0.289 |
| 11–100 | 90 | 110,403 | 246 | 0.223 |
| 101–1000 | 900 | 279,031 | 1,610 | 0.577 |
| 1001+ | 4,514 | 297,765 | 3,693 | 1.240 |
| Outside selection | 4,781 | 31,137 | 0 | 0.000 |
| Measure | Reference model | Native week | Recovered |
|---|---|---|---|
| Mean tone | -0.535 | -0.629 | 0.140 |
| ARMEDCONFLICT (%) | 11.50 | 11.22 | 7.99 |
| MILITARY (%) | 4.99 | 4.29 | 7.52 |
| Iran country mention (%) | 1.69 | 1.26 | 1.17 |
| Precision | Records | Median (days) | 95th (days) |
|---|---|---|---|
| Common Crawl | |||
| Day only | 16,174 | ||
| Minute | 8,829 | 0.048 | 1.124 |
| Wayback | |||
| Day only | 27,981 | ||
| Minute | 228,636 | 0.238 | 2.674 |
| Offset (h) | Later (%) | Outside | TV |
|---|---|---|---|
| 0 | 99.8 | 0 | 0.171 |
| 1 | 75.8 | 875 | 0.164 |
| 2 | 54.0 | 1,787 | 0.161 |
| 4 | 25.6 | 2,375 | 0.180 |
| 6 | 12.7 | 3,165 | 0.211 |
KG Construction
Knowledge graph construction turns text, tables, images, and databases into structured entities and relationships that machines can query and reason over.
See all 207 papers