DescribeEvents/gdelt_news_events

GDELT News Events

Curated geopolitical event sequences from the GDELT 2.0 Event Database, designed for temporal point process (TPP) and marked temporal point process (MTPP) modeling. Each sequence tracks interactions between a country pair (e.g., USA–Russia) over time, where the prediction target is the CAMEO event type of each geopolitical development.

Schema

Each record is a dictionary with 10 fields:

Field	Type	Description
seq_idx	int	Sequence index
seq_len	int	Number of events in the sequence
type_category	str	Dominant QuadClass for the sequence
span_days	float	Time span of the sequence in days
description	str	Natural language description of the country pair and time window
metadata	str (JSON)	country_pair, country_names, date_range_start, date_range_end, span_days, num_event_types
time_since_start	list[float]	Time since the first event (in days)
time_since_last_event	list[float]	Time since the previous event (in days)
type_event	list[str]	CAMEO-derived event type (see below)
type_text	list[str]	Natural language description of each event

Event Types (7 types from 20 CAMEO codes, ~99 base-code verbs)

20 CAMEO root codes are merged into 7 action-based types. Each event uses a specific verb from ~99 CAMEO base codes (3-digit EventBaseCode) for precise action descriptions, falling back to 20 root-level verbs when the base code is unavailable:

type_event	CAMEO codes	Example verbs (base-code level)
public_statement	01, 02	Made optimistic remarks; declined to comment; appealed for aid
consultation	04, 09	Discussed by telephone; made an official visit; engaged in formal negotiations
diplomatic_engagement	03, 05	Signed a formal agreement; formed a coalition; praised or endorsed
material_cooperation	06, 07, 08	Provided economic aid; shared intelligence; de-escalated military engagement
disapproval	10, 11, 12, 13	Criticized or denounced; vetoed proposals; issued an ultimatum
coercion	14, 15, 16, 17	Conducted military exercises; imposed an embargo; arrested or detained individuals
armed_conflict	18, 19, 20	Carried out a bombing; occupied territory; used aerial weapons

Type Categories (QuadClass)

Each event also carries a QuadClass category: Verbal Cooperation, Material Cooperation, Verbal Conflict, or Material Conflict. The sequence-level type_category field stores the dominant category for the sequence.

Curation Filters

Filter	Value	Description
IsRootEvent	1	Only root events (de-duplicated by GDELT)
NumMentions	>= 2	Events covered by multiple sources
same-country	excluded	Only cross-country interactions
acronym-allowlist	~28 acronyms	Explicit set (UN, EU, NATO, ISIS, etc.) kept uppercase; all other ALL-CAPS names title-cased
known-group-override	12 orgs	NATO, UN, EU etc. use proper names via KnownGroupCode (no FIPS leaks)
multi-role-actors	combined	GOV+MIL → "government and military forces"
BUS-actor-filter	excluded	Corporate events excluded (not geopolitics)
MED/EDU-conflict-filter	excluded	Media/educational actors excluded from armed conflict (CAMEO 18-20)
armed-conflict-plausibility	NumMentions >= 3	CAMEO 18-20: mention threshold filters noise (location matching dropped — GDELT geocodes news origin, not conflict location)
A-artifact-suppression	stripped	Short GDELT artifacts like "A US" (article + country code) removed
location-suppressed	removed	GDELT ActionGeo reflects news origin, not event location — all location text removed
daily-aggregation	weighted-mean Goldstein	One event per day per pair (NumMentions-weighted)
min-events	40	Min events per sequence
max-events	100	Max events per sequence
min-types	3	At least 3 distinct event types per sequence
max-dominant-ratio	0.35	No single type exceeds 35% of events
max-repeat-ratio	0.35	No more than 35% consecutive same-type events
max-swing-ratio	0.15	No more than 15% consecutive Goldstein swings >= 12
min-span-days	30	Minimum sequence span
max-span-days	1200	Maximum sequence span

Event Text

Each event's type_text is a concise sentence: actor1 + action verb + actor2. Built from structured GDELT fields (actor names, roles, countries) — no headline scraping. Known organizations (NATO, UN, EU) use proper names via KnownGroupCode lookup, and multi-role actors combine types (e.g. "government and military forces"):

government and military officials from United States engaged in formal negotiations with Russia.

NATO cooperated militarily with United States.

Russia imposed an embargo or sanctions on Ukraine.

Example

{
  "seq_idx": 0,
  "seq_len": 100,
  "type_category": "Verbal Cooperation",
  "span_days": 153.0,
  "description": "Geopolitical event timeline between United Kingdom and United States, 2024...",
  "metadata": "{\"country_pair\": [\"GBR\", \"USA\"], \"country_names\": [\"United Kingdom\", \"United States\"], ...}",
  "time_since_start": [0.0, 1.0, 2.0, ...],
  "time_since_last_event": [0.0, 1.0, 1.0, ...],
  "type_event": ["coercion", "disapproval", "diplomatic_engagement", ...],
  "type_text": ["United States imposed an embargo or sanctions on United Kingdom.", ...]
}

Intended Use

• Training and evaluating temporal point process models
• Studying geopolitical interaction patterns between country pairs
• Benchmarking next-event prediction and event forecasting
• Modeling marked temporal point processes with rich text marks

Citation

If you use this dataset, please cite:

@dataset{gdelt_news_events_2025,
  title={GDELT News Events},
  author={XiaoBB},
  year={2025},
  url={https://huggingface.co/datasets/XiaoBB/gdelt_news_events},
  note={Curated from GDELT 2.0 Event Database, 2022-2024}
}