A Knowledge Graph is a structured Knowledge Base. Knowledge Graphs store facts in the form of relations between different entities. By extracting facts from a knowledge base and representing these facts in the form entities and relations, a knowledge graph claims to have an understanding of the information.
More formally, a Knowledge Graph is a very large semantic nets that integrate various and heterogeneous information sources to represent knowledge about certain domains of discourse.
- A graph is a mathematical structure in which some pairs in a set of objects are somehow related.
- We describe knowledge to the actions of an agent.
- An agent would interpret a knowledge graph to make rational decisions to take actions to reach its goals.
An ontology is an explicit, formal specification of a shared conceptualization.
- conceptualization: abstract model of domain related expressions
- specification: domain related
- explicit: semantics of all expressions is clear
- formal: machine-readable
- shared: concensus
A knowledge graph (i) mainly describes real-world entities and their interrelations, organized in a graph, (ii) defines possible classes and relations of entities in a schema, (iii) allows for potentially interrelating arbitrary entities with each other and (iv) covers various topical domains.
In essence, Resource Description Framework (RDF) is an Information Model, much like the Relational Database Model. In fact, one could say that RDF is a Relational "Triple" Model, in which a statement of is called a triple.
- Expansion: knowledge graphs are incomplete
- data linking (entity resolution, reference reconciliation)
- link prediction: add relations
- ontology matching: connect graphs
- missing values prediction / inference
- Enrichment: new knowledge merges from knowledge graphs
- knowledge discovery: key discovery
- automatic reasoning and planning
- Validation: knowledge graphs may contain errors
- link validation
- dealing with errors and ambiguities
Data linking or identity link detection consist of detecting whether two descriptions of resources refer to the same real-world entity.
Some approaches are available
- Local approaches: consider properties to compare pairs of instances independently
- Global approaches: consider data type properties as well as object properties to propagate similarity scores / linking decisions (collective data linking)
- Supervised approaches: need samples of linked data to learn models, or need interactions with expert (interactive approaches)
- Informed approaches: need knowledge to be declared in the ontology or in other format
Keys restaurant(r1) $\uparrow$ restaurant(r2) $\uparrow$ address(r1, a) $\uparrow$ address(r2, a) -> sameAs(r1, r2)
Disjoint classes C1(x) $\uparrow$ C2(y) -> differentFrom(x, y)
Functional Datatype properties sameAs(r1, r2) $\uparrow$ city(r1, c1) $\uparrow$ city(r2, c2) -> equivalentString(c1, c2)
Local-UNA authored(p, a1) $\uparrow$ authored(p, a2) -> differentFrom(a1, a2)
Referring expression name(p1, 'Obama') $\uparrow$ profession(p1, 'president') -> sameAs(p1, http://...81)