What is a knowledge graph?
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
A knowledge graph is a way of organizing information around things and the relationships between them. The things might be people, products, places or ideas. The relationships specify how they are connected. Here, “graph” means a network of connected items, not a bar chart or a line chart.
Consider a music collection. An artist recorded an album; the album contains a song; another artist wrote that song. Each item is represented as a node, meaning a distinct point in the network. Each connection is an edge, meaning a named relationship between two items. These names make the connections meaningful rather than merely visual.
A simple fact can be written as three parts: “Artist A — recorded — Album B.” This arrangement is often called a triple. Linking many such facts makes it possible to ask questions that cross several connections: Which songs on this album were written by someone other than the performer? The answer comes from following the relevant relationships.
A knowledge graph is not necessarily an AI model. A model is software that learns patterns from examples. A graph can instead contain facts entered by people or drawn from existing records. AI can help propose new connections or extract information from documents, but those additions still need checking.
The distinction from ordinary text is that the relationships are explicit. A paragraph may mention an artist, an album and a song without clearly specifying who wrote what. A graph requires those relationships to be stated. It can also record where a fact came from and when it was valid, which helps when sources disagree or information changes.
Careful identification matters. Two performers with the same name should not accidentally become one person in the graph. Equally, two spellings of the same person’s name should not automatically become separate people. Connections are only useful when the underlying items and claims are represented accurately.
In biomedical research and clinical decision support, a knowledge graph can connect a medication to its active ingredient, target pathway, contraindicated conditions and documented drug–drug interactions. That explicit structure lets an application answer multi-step relational questions that unstructured text search can easily miss—provided entity resolution, source provenance and update rules keep the underlying clinical facts accurate.
Sources: W3C — RDF 1.1 Concepts and Abstract Syntax; Hogan and colleagues — Knowledge Graphs.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy