A knowledge graph is a representation of knowledge regarding a particular domain in machine-readable form
The representation of knowledge in graph form is a key element in the effective, contextual searching for rich information and knowledge, in decision-making, artificial intelligence applications, voice assistants …
A knowledge graph is composed of three elements: an ontology (data model), reference concepts or controlled vocabularies, and the resources covered by the graph.
The Strengths of Knowledge Graphs
- data that are easily accessible and comprehensible because they’re organized according to specific professional perspectives,
- data that are durable and stable, at the service of applications for processing and spreading information,
- resources that are directly readable by artificial intelligence and logical reasoning tools,
- quality data thanks to a range of automated, intelligent tools, dedicated to controlling the graph’s quality and completeness,
- knowledge graphs are easy to enrich, with a negligible impact on the applications that already use it,
- significantly lower implementation and maintenance costs than relational databases,
- native multilingual management,
- an interoperability of data fostered by the management of stable identifiers, the alignment with external resources, and the implementation of shared reference concepts,
- an environment and standards built natively on web standards.
Coverage
of a knowledge graph
In the framework of a project, it isn’t always easy to define a knowledge graph’s coverage. Should it describe all possible details? Should it be as simple as possible?
An effective and reliable method is to canvas future users for all the questions the graph should answer, then build the ontology and supply the graph in order to answer this specific type of questions
Enriching knowledge graphs is simple. At any moment you can enrich the ontology, the controlled vocabularies, and the types of knowledge according to users’ actual needs
Ontology
The modeling of knowledge in the graph is defined by its ontology (data model). The ontology defines the categories of resources (people, works, places, events …), their properties (name, date of birth, textual description …) and the relations they possess (created by, uncorrupted, carried out by …). Ontologists dispose of a series of basic ontologies that describe frequently-used types of modeling: modeling of an event, a document, a work, an organization, a person, a reference concept, modeling of time, of geographic positions …
To create the graph’s ontology, you can combine these basic ontologies and add any missing elements. You can also design an ontology that builds on these basic ontologies.
Depending on the graph’s enrichment needs, you can add new categories of resources, relations, and properties.
Controlled Vocabularies
Knowledge graphs rely on a collection of controlled vocabularies (reference concepts) that make it possible to identify the concepts used to describe the resources (for example: musical genres, types of machines or legal texts, people’s roles in an organization, a list of countries or languages …). The creation of these controlled vocabularies is one of the steps of implementing a knowledge graph.
Supplying
the Knowledge Graph
The information that supplies a knowledge graph can come from a wide variety of sources: relational databases, XML documents, knowledge extracted from texts or images, manual input, other knowledge graphs …
One of the knowledge graph’s strengths is its ability to integrate and link heterogeneous data from multiple sources.
Pour mettre en place un processus d’alimentation automatisée, on réalise un alignement entre la description des ressources à intégrer et l’ontologie du graphe de connaissance, puis on met en place les traitements d’extraction, conversion et alimentation du graphe.
Reasoning
The knowledge graph is designed for reasoning. Reasoning can be based on the rules defined in the ontology or on a set of separately maintained business rules.
Reasoning is used to infer new knowledge from the graph’s knowledge, to complete external data with the graph’s knowledge, and to control, automatically and continually, the knowledge graph’s consistency and completeness.
The Knowledge Graph at the Service of Applications: A Data-Driven Approach
The implementation of a knowledge graph is part of a data-driven organizational strategy in which aggregate, machine-ready knowledge becomes the basis of numerous applications and services, such as:
- the publication of websites to search and navigate in a body of associated knowledge and content,
- the enrichment of a search and navigation tool in order to provide it with context and knowledge,
- the implementation of vocal assistants and chatbots able to dialogue with users,
- the semantic annotation of new content relying on the graph’s knowledge,
- the enrichment of existing applications with data from a specialized knowledge base,
- supplying AI applications with training data
- …
Each knowledge graph becomes the center of an applications ecosystem.