SYSTEM Cited by 3 sources
Pinterest Home Feed¶
Definition¶
Home Feed (HF) is Pinterest's primary content-discovery product surface where users see a personalised feed of Pins and ads. One of three Pinterest ads surfaces (with Search and Related Pins), each with distinct user intent, feature availability, and serving cost profile.
Cascaded funnel architecture¶
Home Feed follows Pinterest's cascaded recommendation funnel (Source: sources/2026-04-07-pinterest-evolution-of-multi-objective-optimization-at-pinterest-home):
The final multi-objective optimization (MOO) / blending layer is architecturally distinct from upstream ranking: it determines the composition of the feed rather than scoring individual Pins. Detailed in systems/pinterest-home-feed-blender.
Pinterest's explicit framing for why HF in particular needs a strong MOO layer: "users tend to have lower intent when visiting Home Feed and their browsing behavior will be significantly impacted by what they see. For example, visually repetitive content is less engaging and is likely to reduce the user's session length and the likelihood that a user will revisit Pinterest."
Role in the unified ads engagement model¶
Home Feed is one of the three surfaces that previously had its own dedicated CTR-prediction model and was unified with Search as the first pair in Pinterest's staged model unification (Source: sources/2026-03-03-pinterest-unifying-ads-engagement-modeling-across-pinterest-surfaces). HF + SR were paired because of their similar CUDA throughput characteristics — substantially cheaper than Related Pins.
In the unified Pinterest Ads Engagement Model, Home Feed has:
- Its own tower tree plus surface-specific modules within the tower tree — handles only HF traffic at serving time.
- Its own view-type-specific calibration — HF traffic calibrated separately from SR traffic because a shared calibration layer "implicitly mixes traffic distributions across surfaces."
- Its own surface-specific checkpoint export from the joint training run.
Caveats¶
- Stub — Home Feed's full architecture (retrieval, ranking beyond ads, organic content mixing) is not ingested. This page is bounded to the Home Feed's role as an ads surface in the 2026-03-03 engagement-modeling post.
- User intent distinction vs SR / RP described in the post only at high level ("different user intent and different feature availability") without concrete feature-set differences.
Seen in¶
- 2026-04-08 Pinterest — Performance for Everyone (sources/2026-04-08-pinterest-performance-for-everyone) — Home Feed named as a "key user experiences' surface" with Visually Complete definition "all the images rendered and videos playing." Canonical example surface in Pinterest's platform-wide automatic UPL measurement built into
BaseSurfaceon Android; extended to iOS and Web. - 2026-04-07 Pinterest — Evolution of Multi-Objective Optimization at Pinterest Home Feed (sources/2026-04-07-pinterest-evolution-of-multi-objective-optimization-at-pinterest-home) — canonical source for the Home Feed cascaded funnel (retrieval → pre-ranking → ranking → multi-objective optimization) and the reasoning for why a dedicated MOO/blending layer matters on a low-intent surface. Details the Blender's V1 (DPP) → V2 (SSD on PyTorch) → V2+ (unified soft-spacing framework) evolution in systems/pinterest-home-feed-blender.
- 2026-03-03 Pinterest — Unifying Ads Engagement Modeling (sources/2026-03-03-pinterest-unifying-ads-engagement-modeling-across-pinterest-surfaces) — canonical; names HF as one of three ads surfaces, paired with SR for first-phase unification.
Related¶
- companies/pinterest
- systems/pinterest-home-feed-blender — the MOO / blending layer feeding the final slate.
- systems/pinterest-ads-engagement-model
- systems/pinterest-search
- systems/pinterest-related-pins
- concepts/feed-diversification · concepts/feed-level-reranking · concepts/retrieval-ranking-funnel — funnel framing.