conceptinterest-graph-recs
Knowledge-based recommender
Notes
Knowledge-based recommender
One-line summary: Recommenders that match elicited requirements or constraint/ontology knowledge — user models may be graphs — rather than a collaborative-filtering interaction matrix.
The insight
Grokipedia distinguishes knowledge-based systems from collaborative filtering and from content-based profile-vector matching. They handle cold-start by asking for requirements instead of waiting for logs; they explain via constraints. User models can be feature–value pairs or graphs of stated goals. The cost is knowledge engineering. A user-drawn interest graph sits closer to this family than to mined knowledge-graph-recommender embeddings. From 2026-08-21-autoresearch-best-architecture-for-a-rec-system-web-app.
Evidence
- From 2026-08-21-autoresearch-best-architecture-for-a-rec-system-web-app: Knowledge-based systems “use structured representations of item features, user requirements, and domain rules … independent of past interaction data.”
- From 2026-08-21-autoresearch-best-architecture-for-a-rec-system-web-app: “User models can also be represented as graphs, leveraging ontology-like structures to capture interconnected preferences.”
- From 2026-08-21-autoresearch-best-architecture-for-a-rec-system-web-app: Advantage vs CF is cold-start; limitation is “knowledge acquisition” and constraint-solving scale.
Contradictions / tensions
- Not the same object as knowledge-graph-recommender (mined item/entity KG as side information).
- Not tested against YouTube’s live feed. See can-a-user-owned-interest-graph-produce-better-recs.
Open questions
- what-web-app-architecture-fits-a-user-owned-interest-graph-recommender
- can-a-user-owned-interest-graph-produce-better-recs
Related
Referenced by