Reasoner

The fundamental operation of applying a set of inference rules to an ABox (assertional facts) and a TBox (terminological axioms) to materialize all entailed logical consequences. This process expands the knowledge graph with implicit triples that were not originally asserted.

• Forward-chaining: Starts from known facts and applies rules to generate new conclusions until saturation.
• Materialization: The physical storage of inferred statements, trading storage space for faster query-time performance.
• Saturation: The state where no new non-redundant statements can be derived from the current rule set and data.

The process of verifying that an ontology and its instance data contain no logical contradictions. A reasoner detects unsatisfiable classes and incoherent models by identifying violations of disjointness axioms, domain/range restrictions, and cardinality constraints.

• Unsatisfiable Class: A class that cannot possibly have any instances, often indicating a modeling error.
• Incoherence: A state where the ontology's schema itself contains contradictory class definitions.
• Explanation Generation: Providing a minimal set of axioms that justify the detected inconsistency for debugging.

The automatic computation of the complete class hierarchy based on the necessary and sufficient conditions defined in the ontology. The reasoner determines all implicit subclass-superclass relationships between named classes.

• Subsumption Checking: Determining if Class A is a subclass of Class B based on their logical definitions.
• Taxonomy Construction: Building the inferred polyhierarchy, which often reveals multiple parentage not explicitly modeled by the ontologist.
• Query Rewriting: Using the inferred hierarchy to expand queries, ensuring that instances of subclasses are returned when querying a superclass.

The task of computing the most specific named classes that an individual belongs to, given its asserted property values and class memberships. This is the inverse of classification, operating on the ABox level.

• Type Inference: Dynamically determining that a specific entity (e.g., 'Patient_001') is an instance of a high-risk cohort based on clinical findings.
• Dynamic Categorization: Assigning individuals to categories defined by complex logical expressions without manual tagging.
• Triggering Rules: Using realized types to activate downstream business logic or clinical decision support alerts.

Leveraging the inferred knowledge to provide complete and correct answers to structured queries. A reasoner rewrites a SPARQL or Cypher query into an expanded form that accounts for all entailed facts, ensuring no implicit data is missed.

• Query Reformulation: Transforming a user's query using ontological axioms to capture semantic equivalents.
• Entailment Regime: The formal specification (e.g., RDFS, OWL 2 RL) that defines which inferred triples must be considered when evaluating a query.
• Virtual ABox: Answering queries by combining explicit data with on-the-fly inference without full materialization.

The selection of a specific computational subset of OWL 2 with favorable computational properties. Profiles like OWL 2 RL, OWL 2 EL, and OWL 2 QL trade expressivity for tractability, enabling reasoning over large-scale clinical datasets.

• OWL 2 EL: Optimized for large biomedical ontologies like SNOMED CT, supporting classification in polynomial time.
• OWL 2 RL: Amenable to implementation using rule-based forward-chaining engines on RDF triplestores.
• Datalog: A declarative logic programming language often used as the underlying formalism for implementing scalable deductive rules.