Data Transformation

This process establishes semantic correspondences between disparate source schemas and a target ontology. It defines how source fields map to ontological classes and properties, resolving structural and naming heterogeneity.

• Key Activity: Defining equivalence (owl:equivalentClass), subsumption (rdfs:subClassOf), and other relationships between concepts.
• Tool Example: The R2RML and RML languages provide standardized ways to declaratively map relational and non-relational data to RDF.
• Outcome: Creates a unified logical view, enabling disparate datasets to be integrated under a common semantic model.

The process of determining whether records from different sources refer to the same real-world entity, then merging them into a single, canonical node in the knowledge graph.

• Core Challenge: Disambiguating entities based on ambiguous or conflicting attributes (e.g., 'J. Smith' vs 'John Smith Corp.').
• Techniques: Employ fuzzy matching (Levenshtein distance), rule-based matching, and machine learning models to compute similarity scores.
• Result: Eliminates duplicate nodes, creating a golden record for each unique entity, which is critical for graph consistency and accurate relationship inference.

The process of detecting and correcting errors, inconsistencies, and invalid values in source data to ensure data quality before integration.

• Common Operations: Standardizing date formats (MM/DD/YYYY → YYYY-MM-DD), correcting misspellings, validating against controlled vocabularies, and handling null values.
• Canonicalization: Converting data with multiple representations into a single, standard authoritative form (e.g., 'USA', 'U.S.A.', 'United States' → http://dbpedia.org/resource/United_States).
• Impact: Directly improves the accuracy of downstream reasoning, querying, and analytics by removing noise and ambiguity.

The process of augmenting raw data with additional attributes, classifications, or links from external sources to increase its informational value and connectability within the graph.

• Methods: Linking entities to authoritative knowledge bases (e.g., DBpedia, Wikidata), adding geospatial coordinates, inferring sentiment from text, or classifying products using a standard taxonomy.
• Semantic Layer: Enrichment often populates the ABox (assertional knowledge) with new facts derived from the TBox (terminological knowledge) or external APIs.
• Benefit: Transforms sparse records into richly connected nodes, enabling more powerful semantic queries and inference.

The technical execution of transformation rules that convert structured source data (CSV, JSON, SQL) into RDF triples conforming to the target ontology.

• Process: Uses mapping languages (R2RML, RML, SPARQL-Generate) to specify how each source field generates a subject, predicate, and object.
• Population: This is the load phase of a Semantic ETL pipeline, where transformed triples are inserted into a triplestore or graph database.
• Output: A populated knowledge graph where instance data (individuals) is explicitly linked according to the defined ontological schema.

The overarching practice of standardizing data from multiple sources to resolve syntactic, structural, and semantic differences, creating a unified, consistent dataset.

• Scope: Broader than cleansing; it ensures all integrated data adheres to the same units of measure, code lists, and business rules.
• Normalization: In a graph context, this involves structuring properties to avoid redundancy (e.g., ensuring 'manufacturer' is a relationship to a Company node, not a repeated string property).
• Goal: To produce a coherent information asset where queries and analytics yield consistent results regardless of the original data source.

A data integration methodology that applies semantic technologies to transform heterogeneous data into a knowledge graph. Unlike traditional ETL, it uses ontologies and RDF mapping languages (like RML) to define the target structure, ensuring the output data carries explicit, machine-interpretable meaning.

• Key Output: RDF triplestore or labeled property graph.
• Core Activity: Mapping source fields to ontology classes and properties.
• Benefit: Creates a semantically interoperable dataset ready for reasoning and complex querying.

A declarative, vendor-neutral language for defining rules to transform heterogeneous data (CSV, JSON, XML, databases) into RDF triples. It extends the W3C's R2RML standard to work with non-relational sources.

• Function: Specifies how to generate subject URIs, predicates, and object values from source data.
• Use Case: The core engine for knowledge graph population, automating the creation of instance data (ABox) from operational systems.
• Example Rule: Maps a customer_id column to a foaf:Person instance with a schema:email property.

The process of converting data that has multiple possible representations into a single, standard, authoritative form (the canonical form). This is critical for consistent entity identity and property values in a knowledge graph.

• Applies To: Dates (ISO 8601), units of measurement, currency codes, product SKUs.
• Example: Transforming "New York City," "NYC," and "New York, NY" into a single URI dbpedia:New_York_City.
• Purpose: Eliminates syntactic variance to enable accurate entity linking and deduplication.

The process of establishing semantic correspondences between the attributes, tables, or classes of two or more heterogeneous data schemas. It defines how source data structures map to a target ontology.

• Output: A set of mappings (e.g., Source.CustomerName ≡ TargetSchema:Person/foaf:name).
• Challenge: Resolving structural and naming differences (e.g., cust_id vs. customerIdentifier).
• Tooling: Often uses semi-automated ontology matching algorithms to propose candidate alignments.

The comprehensive process of standardizing data from disparate sources by resolving syntactic, structural, and semantic differences to create a unified, consistent dataset. It encompasses transformation, cleansing, and alignment.

• Scope: Broader than simple format conversion; it ensures integrated data can be used coherently.
• Steps: Often includes normalization, canonicalization, unit conversion, and code list mapping.
• Goal: To produce a coherent single source of truth from multiple, conflicting sources.

The specific process of extracting, transforming, and loading instance data (ABox assertions) from source systems into the structure defined by an ontology (TBox). It is the execution phase of a semantic ETL pipeline.

• Input: Raw source data + Ontology + Mapping Rules (e.g., RML).
• Activity: Creates concrete nodes (entities) and edges (relationships) in the graph database.
• Quality Gate: Requires validation against ontological constraints (domain/range, cardinality) to ensure logical consistency.