GDELT: Global Database of Events & Tone

• GDELT is a comprehensive global database that structures news events, actor interactions, sentiment, and metadata across granular spatiotemporal scales.
• It employs advanced event coding, language metadata, and affect analysis techniques to support computational social science and machine learning applications.
• The platform integrates schema-driven relational interfaces with knowledge graph paradigms, enabling efficient event detection, forecasting, and data-driven insights.

The Global Database of Events, Language, and Tone (GDELT) is a massive, multi-modal, continually updated repository of structured representations of news media content, capturing global events, actor relations, linguistic sentiment, and associated metadata at granular spatiotemporal scales. It is designed to support computational social science, event detection, forecasting, and knowledge extraction across disciplines by providing exhaustive, high-frequency coverage of the world’s news cycle in machine-actionable form. Its architecture integrates event coding, language metadata, affect and sentiment analysis, and entity resolution, exposing these data through both schema-driven relational interfaces and knowledge graph paradigms suitable for downstream machine learning, evaluation, and retrieval-augmented generation workflows.

1. Data Architecture and Core Schema

GDELT’s data model is partitioned into several interlinked tables reflecting different levels of semantic annotation:

• Events Table (e.g., expert.csv): Each row encodes a single event, indexed by GLOBALEVENTID. Key fields comprise EventDate, Actor1Code/Actor2Code, standardized country/organization codes, EventCode (following the CAMEO ontology), GoldsteinScale (quantifying perceived impact), NumArticles/NumMentions/NumSources (measures of media visibility), AvgTone (average sentiment), and geolocational data at various granularities (Myers et al., 10 Mar 2025).
• Global Knowledge Graph (GKG): At the document/article level, this contains fields such as GKGRECORDID, DATE, SourceCommonName, V2Themes (thematic labels), V2Locations, V2Persons, V2Organizations, V2Tone (fine-grained affect vector), and GCAM (thousands of soft counts from four sentiment lexica and topical dictionaries) (Tilly et al., 2020).
• Mentions Table: Links events and articles, assigning each mention confidence, mention-level tone, and locational context. It provides foreign-key joins to both Events and GKG tables, enabling fusion of document-level and event-level semantics [250