Modèle sémantique, graphe de connaissances, ontologie
Three ways of writing the meaning of your data outside the heads of people, each answering a different question. The semantic model says how to read and compute: what revenue, margin or an active customer are, and by which dimensions they get sliced. The knowledge graph says what is connected to what: this order belongs to this customer, contains these products, ships from this warehouse. The ontology says what those objects and links mean for the business and what is allowed: which states an order can be in, which actions move it from one to the next, and who may trigger them.
Strengths
- Gives a clean criterion to choose: calculation, relationship or rule, each has its layer
- Moves business knowledge out of queries and code so any agent, dashboard or application can read it
- Cuts the definitions an agent makes up: it reads what a term means instead of guessing
Limitations
- The word "ontology" is also a sales pitch: check whether it really covers rules and actions, or just a renamed graph
- Each layer needs upkeep: a stale definition in the model misleads the agent more confidently than a missing one
Best for
- PMs and POs scoping an agent wired to company data who must say what it needs to know before what it needs to do
- Data teams seeing the same numbers computed three times three ways in three tools