Entity disambiguation: beyond recognition PHASE 1

Named entity recognition (NER) identifies mentions of people, organisations, places, and relevant concepts within unstructured texts using natural language processing techniques. Once these mentions are detected, disambiguation determines which specific entity is being referred to, distinguishing between homonyms, linking it to unique identifiers, and connecting it to its corresponding regulatory and ontological context.

Argos resolves ambiguity through four integrated mechanisms that transform textual mentions into unambiguous references to structured entities:

The system processes the surrounding syntactic and semantic structures to infer the correct identity of each entity. Detecting a term is not enough: it is necessary to determine which specific referent it points to within its context of use.

Contextual signals include syntactic position, adjacent terms, thematic markers, and cross-references within the document. The system evaluates these signals in combination to resolve cases where the same term designates different entities depending on the discursive context.

Entities are not matched against text strings but against structured knowledge representations. Each disambiguation links to canonical resources that ensure semantic consistency across document collections.

Each entity resolution carries a probabilistic confidence score derived from contextual signals, ontological coherence, and corpus validation. The system quantifies the certainty of each disambiguation.

Entities evolve: information is amended, organisations are restructured, frameworks change. The system maintains temporal validity, recognising that relationships between entities vary according to the point in time being referenced. Each entity is associated with temporal validity ranges.

Recognised and disambiguated entities do not remain as isolated data points. Argos maps them to formal ontological structures that provide a shared semantic framework independent of the application domain.

This mapping is not taxonomic window dressing. It enables three fundamental capabilities that transform scattered data into queryable intelligence.

Documents from different sources, languages, or structures become queryable within a unified semantic framework. Terminological differences are resolved through mapping to common ontological classes. Ontologies act as a lingua franca that enables queries independent of the specific terminology of each domain.

Ontological mapping enables the derivation of implicit knowledge through logical rules. If ontological classes establish that certain types of entities maintain specific relationships, the system infers those relationships even when they are not explicitly declared. Inference rules operate on classes and properties, allowing derived knowledge to scale across the entire corpus.

Ontologies connect entities with geographical features, temporal frameworks, and organisational hierarchies. A single entity becomes a nexus linking multiple dimensions of knowledge, enabling queries that cut across traditionally separate domains.

Basic recognition provides minimal information: an entity exists and was mentioned in the text. Enrichment transforms sparse data into queryable intelligence infrastructure.

For each entity, Argos automatically retrieves and structures information from canonical repositories. Attributes are not stored as free text but as structured properties linked to controlled vocabularies. Relational context is expressed through typed relationships in the graph.

Authoritative attributes include official identifiers, registration codes, thematic classifications, and supervisory relationships, extracted from domain canonical repositories. Relational context maps hierarchical structures, functional relationships, and temporal validity ranges. Temporal metadata tracks creation dates, modification history, and validity periods. Multilingual consistency maintains semantic identity across languages through properties that ensure queries in any language retrieve the same underlying entities.

Argos does not produce annotated documents. It produces queryable knowledge graphs expressed in RDF triples, where each fact is represented as subject-predicate-object.

The graph enables four fundamental capabilities through standard semantic web protocols.

Access through SPARQL queries allows complex queries to be expressed that combine thematic, temporal, and relational criteria. Graph traversal turns relationships between entities into navigable paths: from a single entity, connected information, related cases, dependencies, and precedents can be discovered. Logical inference via ontologies enables reasoning engines to derive implicit knowledge. Federated integration allows RDF graphs to be composed: domain-specific intelligence integrates with other knowledge graphs, databases, and information systems.