Hypothesis-Driven Shelf Generation for Personalised Recommendation

2026-07-28

AI Summary

This work addresses the limitations of traditional recommendation systems, which rely on handcrafted templates and struggle to capture users’ long-tail interests. The authors propose a natural language hypothesis-driven framework for personalized shelf generation that decouples shelf planning from content retrieval. The approach comprises four stages: hypothesis generation, catalog satisfaction, shelf alignment, and offline evaluation. For the first time, large language models (LLMs) are leveraged to generate semantic hypotheses, integrated with generative retrieval, candidate selection, and an LLM-as-a-judge evaluation mechanism, enabling independent optimization of planning and retrieval. This method substantially expands the scope of personalized content provisioning and achieves user engagement on par with strong baselines in certain scenarios.

Abstract

Modern recommendation interfaces organise content into shelves: themed rows such as "More of What You Like" or "New Releases for You." In production systems, these shelves are typically defined through hand-crafted templates coupled with dedicated retrieval logic. While effective for broad recommendation intents, this approach does not scale to the long tail of individual taste. We present a content-hypothesis-driven shelf generation system for Spotify Home that replaces fixed templates with natural-language hypotheses describing what a personalised shelf should contain. The system has four stages hypothesis generation, catalogue fulfilment, shelf alignment, and offline serving. This decomposition decouples shelf planning from catalogue fulfilment, supports independent optimisation of planning and retrieval, and enables both constrained generative retrieval over catalogue entities and distillation of frontier LLM behaviour into compact models. Our production pipeline combines hypothesis generation, generative retrieval, candidate selection and shelf alignment, offline LLM-as-a-judge evaluation, and precomputed serving. We describe the end-to-end architecture and evaluate it through offline analyses and an early online evaluation under uniform random exposure on Spotify Home. Results show that hypothesis-driven shelves substantially expand personalised recommendation supply with engagement that varies by content type and is competitive with strong existing shelves in some settings.

Problem

• personalised recommendation
• shelf generation
• long-tail preferences
• recommendation interfaces
• content organisation

Innovation

• hypothesis-driven recommendation
• generative retrieval
• personalized shelves
• LLM distillation
• recommendation system architecture