Semantic Triples

Semantic triples are formalized through the Resource Description Framework (RDF), a W3C standard that provides the abstract syntax for representing metadata about web resources. RDF serializations include:

• Turtle (.ttl): Human-readable compact syntax
• JSON-LD: JSON-based format for web developers
• RDF/XML: XML-based syntax for legacy systems
• N-Triples: Line-based format for streaming

Each triple forms a directed edge in a graph, where subjects and objects are nodes and predicates are labeled, directed arcs connecting them.

A single, atomic unit of knowledge represented in a subject-predicate-object structure. Each assertion encodes one fact about an entity, such as 'Tesla' - 'manufactures' - 'Model 3'. These assertions are the indivisible building blocks of knowledge graphs, enabling machines to parse and reason over factual statements. A collection of triple assertions forms a triplestore, which can be queried using languages like SPARQL to infer new relationships and answer complex questions.

The NLP task of identifying a textual mention of an entity and resolving it to its unique, unambiguous entry in a target knowledge base. For example, distinguishing whether 'Paris' refers to the city in France or the socialite Paris Hilton based on surrounding context. Entity linking is critical for disambiguation in search engines and AI models, ensuring that semantic triples connect to the correct node in a knowledge graph rather than creating false associations.

The process of determining semantic correspondences between concepts from two different ontologies to enable data interoperability. When one knowledge base uses 'CEO' and another uses 'Chief Executive Officer', ontology alignment establishes that these terms are equivalent classes. This is essential for merging enterprise data with public knowledge graphs like Wikidata or DBpedia, ensuring that semantic triples from disparate systems can be unified under a consistent conceptual framework.

A machine learning technique that transforms the nodes and edges of a knowledge graph into low-dimensional, continuous vector representations. These embeddings preserve the graph's structural and relational properties, allowing semantic triples to be processed by neural networks. Techniques like TransE and RotatE learn to position entities in vector space such that the relationship 'foundedBy' is encoded as a consistent translation vector between a company node and a person node.

A fundamental NLP task that locates and classifies named entities in unstructured text into pre-defined categories such as person, organization, location, or brand. NER is the prerequisite step for constructing semantic triples from raw documents. Modern NER systems use transformer-based models like BERT to achieve high accuracy, identifying not just the entity span but also its type, which determines the valid predicates that can connect it within a knowledge graph.

A scoring metric that quantifies the contextual importance of a specific named entity within a document relative to all other entities mentioned. A high salience score indicates the entity is central to the document's topic, not merely a passing reference. AI models use salience to determine which semantic triples to prioritize when generating summaries or answering queries, ensuring that the most relevant facts about a brand or concept surface first.