Overview#
Raw text contains intelligence. A financial record references an IBAN and a company name. An incident report names a location, a vehicle, and two individuals. A communications intercept links an alias to a known entity. Extracting that structured information manually is slow and error-prone. The AI Entity Extraction platform does it automatically, at scale, across dozens of languages.
The system recognises a wide range of entity types including people, organisations, locations, dates, financial amounts, and domain-specific identifiers, then resolves variant mentions to canonical forms and discovers relationships between entities to build structured knowledge from raw documents. Extracted data maps directly onto the POLE model (Person, Organisation, Location, Object, Event) used throughout Argus's 432 business domains.
Key Features#
-
Named Entity Recognition: Identifies and classifies entities across standard types (person, organisation, location, date, monetary amount) and domain-specific types (bank accounts, case numbers, regulatory references, digital addresses, contact information).
-
Entity Resolution and Disambiguation: Resolves different mentions of the same entity to a single canonical form using contextual analysis, semantic similarity, and knowledge base linking.
-
Relationship Extraction: Discovers and classifies relationships between entities including ownership, employment, transactions, family connections, and location associations.
-
Multi-Language Support: Full entity recognition across dozens of languages with script recognition and encoding detection.
-
Nested and Implicit Entity Detection: Recognises entities embedded within other entities and infers entities from contextual clues even when not explicitly mentioned.
-
Confidence Scoring: Provides extraction, resolution, and relationship confidence scores that enable automatic processing of high-confidence results while flagging uncertain results for human review.
-
Entity Network Graph Construction: Builds queryable knowledge graphs with entities as nodes and relationships as edges, enabling link analysis and network visualisation.
-
Format Standardisation: Normalises extracted entities to standard representations for consistent downstream processing.
Use Cases#
-
Compliance and AML/KYC Screening: Automatically extract parties, accounts, and relationships from documents for regulatory monitoring and screening against watchlists. Financial crime units at banks and payment processors reduce manual review time while improving screening coverage.
-
M&A Due Diligence: Identify key parties, obligations, risks, and relationships across hundreds of contracts and filings in days rather than weeks.
-
Intelligence Analysis: Map entity networks from unstructured sources including news, communications, and public records to discover hidden connections and support link analysis. Intelligence agencies and law enforcement organisations apply this to build actionable network graphs from raw source material.
-
Fraud Detection: Extract entities and transaction patterns from documents to identify anomalies, match against known fraud patterns, and build evidence chains.
-
Data Enrichment: Automatically extract structured metadata from unstructured documents to populate databases, verify records, and link related information across sources.
Integration#
The platform integrates with document processing pipelines, knowledge bases, case management systems, and graph databases. Extracted entities and relationships can be exported as structured data or visualised as network graphs for investigative analysis.
Open Standards#
- OASIS STIX 2.1: Extracted entities and relationships are mapped to and from STIX 2.1 Structured Threat Information Expression objects (threat-actor, indicator, attack-pattern, malware, vulnerability, relationship SDOs), enabling bidirectional exchange with compatible threat-intelligence platforms.
- OASIS TAXII 2.1: Automated polling of TAXII 2.1 feeds ingests STIX bundles whose parsed objects feed directly into the entity extraction and resolution pipeline.
- FIRST Traffic Light Protocol (TLP): Extracted entities inherit TLP classification markings (WHITE/GREEN/AMBER/RED/CLEAR/AMBER+STRICT) sourced from STIX object_marking_refs, governing downstream access and sharing permissions.
- POLE Model (Person, Organisation, Location, Event/Object): All extracted entities are normalised onto the POLE schema used by UK policing and intelligence standards, providing a consistent canonical form across all 432 business domains.
- GraphQL (June 2018 Specification): The full entity extraction API, including named entity recognition requests, relationship queries, entity merging, and confidence-scored suggestions, is exposed exclusively through a GraphQL interface.
- GLEIF LEI (ISO 17442): Legal Entity Identifiers from the Global Legal Entity Identifier Foundation are ingested and matched during organisation entity resolution, providing verified canonical identities for corporate entities.
- Exif (JEITA/CIPA DCF Standard): EXIF metadata is extracted from image evidence to derive geolocation coordinates, capture timestamps, and device attributes that seed location and object entity records.
Last Reviewed: 2026-02-05 Last Updated: 2026-04-14