SYSTEM Cited by 1 source
Pinterest OmniSage¶
Definition¶
OmniSage is Pinterest's large-scale, multi-entity heterogeneous graph representation learning system (KDD 2025). It advances the Pinterest embedding lineage by fusing visual/semantic features, interaction-graph signals, and Pin-Board topology into a unified representation (Source: sources/2026-07-27-pinterest-pinner-progression-better-use-case-representation-driving-weekly-active-user-growth).
Why it matters for UIC¶
The critical property of OmniSage for User Interest Clusters is that "closeness" in the embedding space encodes functional utility as well as visual similarity. A hiking boot and a trail-mix bar are neighbors because they are co-curated on "Hiking Trip" boards and co-engaged by the same users — not because they look alike.
Consequence: clustering a user's engaged Pins in OmniSage space produces clusters corresponding to coherent life projects (planning a camping trip, redecorating a bedroom) rather than visual categories. This is what makes UIC's use-case representation meaningful — use-cases are functional, not visual.
Lineage¶
- PinnerSage — multi-embedding user representation via clustering engagement history (KDD 2020).
- OmniSage — multi-entity graph representation (KDD 2025); the embedding space UICs are built in.
- Related graph-embedding heritage: GraphSage.
Seen in¶
- sources/2026-07-27-pinterest-pinner-progression-better-use-case-representation-driving-weekly-active-user-growth — as the embedding substrate for UIC medoids and landmark Pins.