Signals & Sources

Not every link between an entity and a topic is created the same way. The graph records where each link came from (the source) and assigns a strength so recommendations and UI can weight and explain them correctly.

Relationship Sources

Every entity-to-ontology link carries a source label. Multiple links from the same entity to the same topic can coexist — for example, a manual interest and a course-completion–inferred interest — so no signal is overwritten.

Source	Typical strength	Meaning
manual	1.0	Set by a human in the mapping editor or profile
catalog	0.7	Set by the AI Librarian after user confirmation
resource_save	0.3	Inferred when a resource is saved to a library
resource_complete	0.6	Inferred when a resource is marked completed
course_enroll	0.4	Inferred when a learner enrolls in a course
course_complete	0.8	Inferred when a learner completes a course
campus_join	0.2	Inferred when a learner joins a campus
session_attend	0.3	Inferred when a learner attends a session
task_complete	0.5	Inferred when a learner completes a task
child_content	0.5 × decay	Propagated from a child entity’s DEVELOPS/FOCUSES_ON
member_activity	0.15	Aggregated from member actions within the entity

Strength is derived from the source and is not user-editable. Users manage interests by adding or removing links, not by changing weights.

Why separate sources?

Recommendation and “My Map” UIs can group and filter by source — e.g. “From Learning Activities” vs “From Library” — and allow users to dismiss inferred mappings without touching manual selections. Deletion can be scoped by source via the API.

Via-Entity Traceability

Activity-inferred links can store a via_entity reference (the platform entity that triggered the inference). For example, an INTERESTED_IN edge from completing a course carries via_entity: course_id. The UI can then explain “Why this topic?” (e.g. “From Course: Python 101”) and support actions like “Remove all topics inferred from this course”.

Behavioral Edges (Profile → Entity)

In addition to INTERESTED_IN (Profile → Topic/Skill), the graph maintains behavioral edges from profiles to platform entities. These support collaborative filtering and provenance:

Relationship	Meaning
ENROLLED_IN	Profile → Course (enrollment)
MEMBER_OF	Profile → Campus (membership)
ATTENDED	Profile → Session (attendance)
SAVED	Profile → Resource (saved to library)
COMPLETED	Profile → Resource, Course, or Task (completion)

Traversals like “users who enrolled in the same courses as me” use these edges. Inferred INTERESTED_IN edges are created when these actions occur, using the topic mappings of the entity (e.g. a course’s DEVELOPS topics) and the appropriate source and strength.

How the Graph Grows From Signals

The graph grows from several mechanisms:

Manual curation — educators and learners add or remove topic mappings; manual links have the highest trust.
AI cataloging — the Librarian proposes mappings; committed links use the catalog source.
Resource save — saving a resource to a library infers links from the resource’s topics to the saving entity (e.g. profile, course) with source resource_save.
Activity hooks — completion, enrollment, attendance, and similar events create behavioral edges and inferred INTERESTED_IN (and, where applicable, member_activity on entities).

Together, these keep the graph aligned with both explicit choices and real usage.

Next Steps

Entity Connections — DEVELOPS, FOCUSES_ON, INTERESTED_IN, and CONTAINS
Recommendations — how strength is used in scoring
AI Cataloging — how the Librarian sets catalog mappings