Is a fuzzy relation a fuzzy set?

Yes. It is a fuzzy set on a Cartesian product, such as the set of ordered pairs from two underlying sets.

Must each row add to one?

No. Membership grades are not automatically transition probabilities or normalized weights.

What is a fuzzy similarity relation?

In a common formal sense, it is a reflexive, symmetric and appropriately transitive fuzzy relation. Informal uses of similarity may impose fewer properties.

AI Glossary / Graphs and symbolic AI / Intermediate

Fuzzy relation

Definition

A fuzzy relation assigns a membership grade to each relevant combination of objects. A binary fuzzy relation, for example, assigns a value between zero and one to each ordered pair, describing how strongly the pair satisfies a specified relationship.

What it means

An ordinary relation says whether one object is related to another. A fuzzy relation can describe degrees of compatibility, resemblance or preference. The pair product-and-requirement might receive a grade indicating how well the product meets that requirement.

The word relation does not imply causation, probability or even symmetry. A product's suitability for a task is different from the task's suitability for the product. The meaning and direction must be defined before the grades are useful.

Relations can involve more than two objects, but binary relations are particularly easy to display as matrices. Rows identify the first object, columns the second, and each cell stores the pair's membership grade.

How it works

Choose the underlying sets and define what membership in the relation means. For finite sets, collect the grades into a matrix. Decide which mathematical properties are required by the application rather than assuming all relations have them.

A relation on one set can be reflexive, assigning one to every self-pair; symmetric, giving the same grade in both directions; or transitive under a specified fuzzy composition rule. A generic fuzzy relation need satisfy none of these special conditions.

Relations can also be composed through intermediate objects. Max-min composition takes the smaller grade along each two-link connection, then chooses the largest result over all intermediate objects. Other compositions use other operators. The composed grade has meaning only when the chosen operations fit the intended relationship and the underlying scales.

Three practical examples

A catalog matches products to requirements with graded suitability rather than a single accepted-or-rejected indicator.
A clustering study represents pairwise similarity and checks which relational properties its similarity measure actually satisfies.
An educational prototype connects topics to skills and skills to learning activities, exploring indirect associations through a declared composition method.

Think of it this way

Think of a table showing how well different plugs fit different sockets. Each row-column pair has its own compatibility description, rather than one score attached to the plug alone.

The analogy illustrates pair-based information. In actual electrical connections, graded physical fit is not permission to use an incompatible device; the model's mathematical grade cannot replace mandatory technical requirements.

Worked example

Composing relations through two intermediates

Consider an invented relation R from a learning goal G to skills S1 and S2: R(G, S1) = 0.8 and R(G, S2) = 0.4. A second relation Q connects those skills to activity A: Q(S1, A) = 0.5 and Q(S2, A) = 0.9.

Under max-min composition, the connection through S1 contributes min(0.8, 0.5) = 0.5. The connection through S2 contributes min(0.4, 0.9) = 0.4. Taking the maximum yields a composed grade of 0.5 between G and A.

If max-product composition were specified instead, the path contributions would be 0.8 × 0.5 = 0.4 and 0.4 × 0.9 = 0.36, giving 0.4 overall. The differing result is caused by the operator choice, not by an arithmetic disagreement.

Neither score is automatically the probability that a learner will achieve the goal. Nor does composition prove that the activity causes learning. Those interpretations would need additional evidence and a suitable model.

Illustrative case study

The fictional Cedar Learning team creates a topic-to-skill matrix for an internal training catalog. It combines this with a skill-to-activity matrix to suggest activities for topics.

Reviewers notice that one matrix describes how strongly a skill is required, while the other describes how often an activity mentions that skill. The numbers share the same range but do not express matching concepts. A high composed score could therefore recommend an activity that names a skill without teaching it.

Cedar revises the second relation to express a documented teaching-coverage judgment and presents suggestions for human review. It records the composition rule alongside the matrices. The case illustrates semantic alignment; it does not claim measured learning improvement.

Common misconceptions and mistakes

Every fuzzy relation is symmetric. Directional preference and suitability relations commonly need not be.
Any similarity score automatically forms a fuzzy equivalence relation. Reflexivity, symmetry and the chosen transitivity condition require checking.
Ordinary matrix multiplication and max-min composition are interchangeable. They use different operations and produce different results.
Grades from unrelated scales can be composed meaningfully just because both lie between zero and one. Their meanings must be compatible.

Business use

Fuzzy relations are useful when organizations need to represent graded associations explicitly, especially for matching, retrieval and preference models. A matrix can make individual pair judgments reviewable and expose gaps in the data.

Specify whether a zero means incompatibility, absence of evidence or a missing value; those are different situations. Document the grade construction, direction and any required relational properties. When chaining relations, assess whether the resulting association supports the decision being made rather than assuming that a mathematically valid calculation is operationally meaningful.

Check your understanding

Answer: A relation gives R(A, B) = 0.9 and R(B, A) = 0.2.

Is it invalid merely because those numbers differ?

Answer: No. A generic fuzzy relation can be directional. It would fail a symmetry requirement if the intended application explicitly required symmetry.

Related terms

Cartesian Product
Membership function (mathematics)
Fuzzy set
membership function
max-min composition
fuzzy similarity
fuzzy ordering
transitivity
relation matrix

Learn next

Add a third intermediate skill to the worked example with grades 0.7 and 0.6. Recalculate both max-min and max-product compositions, then explain which path supplies the final grade in each case and why.

Sources

Zadeh's primary paper defines fuzzy similarity and ordering properties: https://www.sciencedirect.com/science/article/pii/S0020025571800051

The author's institutional archive identifies the preceding technical report: https://www2.eecs.berkeley.edu/Pubs/TechRpts/1970/28668.html

A Berkeley technical report explicitly develops discrete max-min relation composition: https://www2.eecs.berkeley.edu/Pubs/TechRpts/1977/ERL-m-77-41.pdf

Relevant publisher and indexed technical-report text was reviewed. The skill matrices and Cedar Learning case are original teaching examples, not validated educational effectiveness measures.

Previous: Fuzzy number Next: Fuzzy retrieval