conceptinterest-graph-recs
Knowledge-graph recommender
Notes
Knowledge-graph recommender
One-line summary: Peer-reviewed recsys methods that use a knowledge graph as side information — typically a mined item/entity graph, not a user-authored interest map.
The insight
The academic recsys record treats graphs as a way to ease sparsity and cold-start and to explain recommendations. A 2020 survey groups those methods into embedding-based, connection-based, and propagation-based families. That is a different object from a hand-drawn interest graph, and a different question from which graph database stores the edges. From 2026-08-21-academic-research-user-owned-interest-graphs-vs-platform-recs.
Evidence
- From 2026-08-21-academic-research-user-owned-interest-graphs-vs-platform-recs: Guo et al. (TKDE 2020) “group them into three categories, i.e., embedding-based methods, connection-based methods, and propagation-based methods.”
- From 2026-08-21-academic-research-user-owned-interest-graphs-vs-platform-recs: “the graphs in this literature are typically mined item/entity graphs, not user-authored interest maps.”
- From 2026-08-21-academic-research-user-owned-interest-graphs-vs-platform-recs: Oprea et al. (IEEE Access 2025) “uses a property graph model implemented in Neo4j” — “customers and products as nodes, purchases as edges, Jaccard over shared purchases.”
- From 2026-08-21-autoresearch-best-architecture-for-a-rec-system-web-app: KGERA (Scientific Reports) scores a mined six-relation MovieLens item KG at test time (~10 ms/user) and stacks that score with ItemKNN / LightGCN / NCF — the graph can change without retraining. Not a user-drawn map.
- From 2026-08-21-autoresearch-best-architecture-for-a-rec-system-web-app: K-RagRec retrieves item-KG subgraphs, re-ranks them, then projects into an LLM. Vanilla document RAG is called noisy and structure-blind.
Contradictions / tensions
- Graph-DB recs ≠ user-owned interest graph. Oprea is purchase-neighborhood collaborative filtering stored in neo4j. Guo’s survey is KG-as-side-information.
Open questions
Related
Referenced by