Coverage Metric

Schema Coverage measures the breadth of entity types and relationship types defined in the graph's ontology compared to the target domain's conceptual model. It assesses if the graph's vocabulary is sufficiently expressive.

• Key Metric: Ratio of defined classes/properties to expected classes/properties.
• Example: A biomedical knowledge graph should include entity types like Gene, Protein, Disease, and relationships like Gene-encodes-Protein and Protein-associatedWith-Disease. Missing these core types indicates poor schema coverage.
• Impact: Inadequate schema coverage limits the graph's ability to model complex domain semantics, crippling downstream reasoning and querying.

Instance Population, or extensional coverage, quantifies the number of real-world entities (instances) of each type that are actually present in the graph. A rich schema is useless without populated instances.

• Key Metric: Counts or densities of instances per class (e.g., 'contains 50,000 Company entities').
• Benchmarking: Compared against known domain corpora or reference datasets (e.g., DBpedia, domain-specific registries).
• Long-Tail Problem: Graphs often cover prominent entities well but miss the 'long tail' of less common but critical instances, which is a key coverage gap.

Relationship Saturation evaluates the density and variety of factual connections (edges) between entities. It moves beyond mere entity counts to assess how richly interconnected the knowledge is.

• Key Metrics: Average degree (connections per node), relationship type distribution, and the completeness ratio of known facts.
• Example: A product knowledge graph with many Product nodes but sparse compatibleWith or substituteFor relationships has low relationship saturation, limiting its utility for recommendation systems.
• Analysis: Often visualized via degree distribution plots; a healthy graph typically has a power-law-like structure, not a disconnected or overly uniform network.

Attribute Completeness measures the extent to which entities have their expected descriptive properties or attributes filled in. It addresses the depth of information per node.

• Key Metric: For a given entity class, the average percentage of defined attributes that are populated (non-null).
• Critical for Search & Analytics: Missing attributes like foundingDate for Company or dosage for Drug severely hamper filtering, faceted search, and analytical queries.
• Assessment: Requires a per-class specification of mandatory and optional attributes to calculate meaningful scores.

This dimension assesses coverage across time and contextual dimensions. A high-quality graph should represent knowledge across relevant time periods and situational contexts.

• Temporal Coverage: Does the graph contain historical facts, current states, and provenance timestamps? Gaps indicate stale or ahistorical data.
• Contextual/Provenance Coverage: Are facts tagged with their source (e.g., specific report, sensor ID, legal jurisdiction)? Lack of provenance limits trust and auditability.
• Use Case: For supply chain graphs, coverage must span the entire product lifecycle from manufacture to delivery.

The most authoritative coverage assessment uses domain-specific benchmarks or gold-standard datasets. This involves a direct, quantitative comparison against a curated corpus of expected knowledge.

• Methodology: Define a set of competency questions or a reference list of entities and facts. Calculate precision, recall, and F1-score for the graph's ability to answer/contain them.
• Example: In academic research, the coverage of a scholarly knowledge graph is tested against known citation networks like the ACL Anthology.
• Outcome: Provides an objective, absolute measure of coverage against an agreed-upon standard, moving beyond internal metrics.

A quantitative metric that assesses the proportion of known or expected facts, attributes, or entities actually present in a knowledge graph against a defined benchmark or ideal state. While Coverage Metric evaluates breadth across a domain, Completeness Ratio measures depth and saturation for known data points.

• Key Difference: Coverage asks "How much of the domain is represented?" Completeness asks "For the parts we know about, how much data is missing?"
• Example: A product knowledge graph may have high coverage of all product categories (good coverage) but missing weight specifications for 30% of known products (70% completeness ratio for that attribute).

A measure of the expressivity and detail of a knowledge graph's underlying ontology, including the diversity of classes, properties, hierarchies, and constraints it defines. Schema Richness provides the structural framework that enables meaningful coverage measurement.

• Direct Relationship: A rich schema with many defined entity types (e.g., Person, Organization, Project, Patent) allows for a more granular assessment of coverage across each type.
• Without Schema Richness, coverage is reduced to simple node and edge counts, lacking semantic meaning. A graph could have high node count but poor coverage if those nodes only represent a few entity types.

A metric that measures the proportion of entities in a knowledge graph that correctly correspond to their real-world referents, free from misidentification or misrepresentation. Coverage and Entity Accuracy are orthogonal; one measures quantity of representation, the other measures its correctness.

• Critical Interaction: High coverage with low entity accuracy creates a large but misleading graph. High entity accuracy with low coverage creates a small, trusted but incomplete dataset.
• Assessment: Coverage is often measured against a domain taxonomy, while Entity Accuracy requires validation against a Gold Standard or expert verification.

Measures the capability of a knowledge graph to provide complete and accurate results for a given set of representative queries, reflecting its practical utility for applications. Coverage is a strong predictor of Query Answerability.

• Mechanism: If a knowledge graph has poor coverage of Supplier entities and supplies relationships, queries like "Which suppliers provide component X?" will have low answerability, regardless of internal optimization.
• Benchmarking: Query Answerability is tested using a benchmark suite of real-world questions, providing an application-layer validation of structural coverage metrics.

A structural quality metric that assesses the degree of linkage within a knowledge graph, often measured by the size of the largest connected component or the average path length between entities. Coverage and Connectedness address different aspects of graph topology.

• Coverage ensures a wide variety of entity types are present.
• Connectedness ensures those entities are meaningfully interlinked into a coherent network, not isolated islands of data.
• A graph can have high coverage but low connectedness if it contains many entity types but few relationships between them, severely limiting inferential power.

A curated, high-quality reference dataset, created by domain experts, used as a benchmark for training, testing, and evaluating knowledge graphs. A Gold Standard is the authoritative source against which metrics like Coverage, Entity Accuracy, and Completeness are often measured.

• Function: Provides the "ground truth" for quantitative assessment. For coverage, a gold standard may be a canonical domain ontology or a complete snapshot of a reference dataset.
• Creation: Involves manual curation and expert validation, making it a costly but essential resource for rigorous quality assessment pipelines.