Relationships explain how one thing connects to another.
A dolphin is-a mammal. A battery is part-of a phone. A teacher teaches a student. A company owns a subsidiary. A storm can cause flooding. A source can support a claim. These are not interchangeable links.
Quick answer: how should relationships be categorised?
- Type: is-a, part-of, located-in, causes, uses, owns, supports, precedes?
- Direction: does A relate to B differently from B to A?
- Symmetry: if A relates to B, must B relate to A?
- Transitivity: if A relates to B and B to C, does A relate to C?
- Cardinality: one-to-one, one-to-many, many-to-many?
- Time: when is the relationship valid?
- Context: under which role, jurisdiction or system?
- Evidence: what supports the link?
- Confidence: certain, probable, disputed?
- Provenance: where did the relation assertion come from?
This article complements How Classification by Relationships Works. That page explains how relationships can help classify entities; this page classifies the relationship types themselves.
1. Start with the relation, not the connected things
Two identical entity types can participate in very different relationships. Person-to-person can mean parent-of, teaches, employs, advises or knows.
2. Is-a expresses type membership
“A whale is a mammal” connects a narrower class to a broader class.
3. Instance-of differs from is-a
“This whale is an instance of bottlenose whale” connects an individual to a class; is-a usually connects classes.
4. Part-of expresses composition
A wheel is part of a bicycle. It is not a kind of bicycle.
5. Contains is the inverse of part-of
Store inverse relations explicitly or derive them consistently.
6. Located-in expresses spatial containment
A school is located in Singapore; this is not the same as being part of the Singapore government.
7. Adjacent-to expresses proximity
Neighbouring places may be adjacent without one containing the other.
8. Connected-to expresses network linkage
Two stations may be connected even when far apart geographically.
9. Causes is stronger than precedes
Earlier events do not automatically cause later events.
10. Contributes-to represents partial causation
Complex outcomes often have several contributing causes rather than one sufficient cause.
11. Enables differs from causes
A condition can make an event possible without producing it directly.
12. Prevents and inhibits are negative causal relations
They describe reduced probability or blocked transition rather than positive production.
13. Uses connects agent to resource or tool
A process uses energy; a student uses a calculator; a machine uses software.
14. Produces connects process to output
Production should be separated from ownership and possession.
15. Owns is a legal or institutional relation
Ownership differs from control, custody and possession.
16. Controls describes operational power
An entity may control an asset without owning it.
17. Member-of expresses institutional belonging
Membership differs from employment and ownership.
18. Employs and works-for are inverses
Typed inverses make role direction clear.
19. Authored-by connects artefact to creator
Creator, editor, publisher and custodian are different relationships.
20. Supports connects evidence to claim
Support strength and independence should remain separate properties.
21. Contradicts connects competing claims
Contradiction is stronger than mere difference.
22. Resembles expresses similarity
Similarity is usually symmetric, but the degree and feature space matter.
23. Precedes and follows express order
Temporal ordering should not be confused with causation.
24. Depends-on expresses requirement
A downstream task can depend on an upstream approval or resource.
25. Directionality matters
Parent-of is directional; sibling-of is typically symmetric.
26. Symmetry should be explicit
If A is adjacent to B, B is normally adjacent to A. If A owns B, B does not own A.
27. Transitivity should be explicit
Some broader-narrower and containment relations may transit; friendship usually should not.
28. Cardinality constrains structure
A person can have many memberships; a record may have one canonical owner under a particular governance model.
29. Time makes relationships dynamic
Employment, ownership, membership and location can begin and end.
30. Context changes relationship validity
A person can supervise another in one project but not in the wider organisation.
31. Provenance belongs to relationship assertions
Record which source claims that one entity owns, causes or supports another.
32. Confidence can attach to the edge
The entities may be certain while the relation between them remains disputed.
33. Generic “related-to” should be a last resort
Untyped links lose meaning and make reasoning difficult.
34. Relation taxonomies need governance
Duplicate relation names, ambiguous inverses and inconsistent direction create graph failure.
35. AI can extract relationship candidates
Models can infer likely edges from text, but inferred relations should preserve evidence and confidence.
36. Relationship extraction can confuse correlation with causation
Use conservative relation types where evidence does not justify a stronger edge.
37. A practical relationship record
- relationship ID;
- subject ID;
- relation type;
- object ID;
- direction;
- inverse relation;
- symmetry;
- transitivity;
- cardinality;
- valid time;
- context;
- evidence;
- confidence;
- provenance;
- version.
38. Relationship categories should enable reasoning
Good relation types let people and machines traverse knowledge without inventing meaning at every step.
39. Relations are not decoration
They often carry the structure that turns isolated records into a knowledge system.
40. The deeper idea
Categories tell us what things are. Relationships tell us what things mean together.
To categorise a relationship well is to preserve what connects two things, in which direction, under what conditions, and with what evidence.
Final answer
Categorise relationships by relation type, direction, symmetry, transitivity, cardinality, time, context, evidence, confidence and provenance. Keep is-a separate from part-of, sequence separate from causation, ownership separate from control, and use typed relations wherever possible instead of generic “related-to”.
