Part IV.b: Whole Page Generation

From Slates to Complete Recommendation Experiences

From Slates to Whole Pages

Previous section: Optimized individual slates

Slate = [item₁, item₂, item₃, item₄, item₅]

This section: Optimize entire pages with multiple carousels

Page = [
  Carousel₁: "Comedies" → [items],
  Carousel₂: "Based on a book" → [items],
  Carousel₃: "Recently added" → [items],
  ...
]

What is a “Whole Page”?

Real-world recommendation interfaces have multiple carousels:

Homepage:

┌──────────────────────
│ Continue Watching    
│ [🎬 🎬 🎬 🎬 🎬 🎬
├──────────────────────
│ Trending Now         
│ [🎬 🎬 🎬 🎬 🎬 🎬
├──────────────────────
│ Comedies             
│ [🎬 🎬 🎬 🎬 🎬 🎬
├──────────────────────
│ Based on a Book      
│ [🎬 🎬 🎬 🎬 🎬 🎬
└──────────────────────

Key Properties:

  • Multiple themed rows
  • Each row = carousel
  • Items scroll horizontally
  • Carousels scroll vertically
  • 2D grid layout

The Whole Page Optimization Problem

Challenge: Optimize the entire page jointly, not each carousel independently

Why? Carousels compete for:

  • User attention (limited screen space)
  • Items (avoid duplication)
  • Diversity (don’t show similar carousels)

Goal: Maximize total page utility

U[page | user] = relevance + diversity + coverage + coherence

Formalizing Whole Page Optimization

(Ding et al., 2019) formalized the whole page optimization problem:

  • Input: User context, candidate items, business constraints
  • Output: Page layout with multiple carousels
  • Objective: Maximize page-level utility

Ding et al. 2019: Key Contributions

Problem formulation:

Maximize: Σᵢ reward(carouselᵢ | user, page₋ᵢ)

Subject to:
  - No item duplication across carousels
  - Global impression constraints
  - Budget constraints
  - Diversity requirements

Solution: Primal-dual algorithm that:

  1. Maintains dual variables for constraints
  2. Greedily composes page carousel-by-carousel
  3. Updates dual weights based on constraint violations

From Optimization to Modular Design

(Kislinger, 2025) presented a modular framework for multi-carousel recommendations:

Key insight:

  1. Topics as constraints for the optimization
  2. Greedy item selection into carousel
  3. Diversified carousel selection as a result

This allows optimizing different objectives at different stages!

Topic Assignment Methods

How to assign items to topics?

  1. Rule-based: Metadata tags (genre, release date, etc.)
  2. Clustering: Group items by embeddings
  3. LLM-based: Use LLM to categorize items
  4. Editorial: Human-curated
  5. Hybrid: Combine multiple methods

Trade-off: Rigid rules vs. flexible semantic grouping

From 2D Grid to 1D Sequence

Visual layout: Page is a 2D grid

Row 1: [Item₁  Item₂  Item₃  Item₄  Item₅  Item₆ ]
Row 2: [Item₇  Item₈  Item₉  Item₁₀ Item₁₁ Item₁₂]
Row 3: [Item₁₃ Item₁₄ Item₁₅ Item₁₆ Item₁₇ Item₁₈]

Transformer input: Need a 1D sequence!

[Item₁, Item₂, Item₃, ..., Item₁₈]

Problem: Lost spatial information! Which item is in which row/column?

2D Positional Embeddings

Solution: Add 2D position embeddings to preserve grid structure

For each item at position (row, col):

item_embedding = item_emb + row_emb + col_emb

Similar to Image Transformer by (Parmar et al., 2018)

Attention Patterns in 2D Grids

1D sequence attention (traditional):

Item₁ → Item₂ → Item₃ → ... → Item₁₈
  ↓      ↓        ↓             ↓
All items attend to all other items equally

2D grid attention (with position embeddings):

Row 1: Item₁ → Item₂ → Item₃  (strong within-row attention)
         ↓       ↓       ↓
Row 2: Item₄ → Item₅ → Item₆  (weaker across-row attention)

Summary: Key Concepts

  1. From slates to whole pages: Multi-carousel optimization
  2. Ding et al. (2019): Formalized whole page optimization
  3. Kislinger (2025): Modular framework with two-stage approach
  4. 2D positional embeddings: Preserve grid structure in 1D sequence
  5. Topic constraints: Items must match carousel themes

References

Ding, W., Govindaraj, D., & Vishwanathan, S. V. N. (2019). Whole page optimization with global constraints. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 3153–3161. https://doi.org/10.1145/3292500.3330675
Kislinger, J. (2025). Full-page recommender: A modular framework for multi-carousel recommendations. Proceedings of the 19th ACM Conference on Recommender Systems. https://doi.org/10.1145/3705328.3748753
Parmar, N., Vaswani, A., Uszkoreit, J., Kaiser, Ł., Shazeer, N., Ku, A., & Tran, D. (2018). Image transformer. Proceedings of the 35th International Conference on Machine Learning, PMLR, 80, 4055–4064. https://arxiv.org/abs/1802.05751