![[2602.21756] Offline Reasoning for Efficient Recommendation: LLM-Empowered Persona-Profiled Item Indexing](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.21756] Offline Reasoning for Efficient Recommendation: LLM-Empowered Persona-Profiled Item Indexing
Summary
The paper presents Persona4Rec, a novel recommendation framework that utilizes offline reasoning with large language models (LLMs) to create interpretable persona representations for items, enhancing the efficiency and scalability of recommender systems.
Why It Matters
As recommender systems become increasingly integral to user experience in digital platforms, the need for efficient and interpretable models is critical. Persona4Rec addresses the high latency of online inference in LLMs, offering a practical solution that balances performance with user understanding, which is essential for real-world applications.
Key Takeaways
- Persona4Rec employs offline reasoning to enhance recommendation efficiency.
- It creates interpretable persona representations from item reviews.
- The framework reduces inference time while maintaining performance.
- Aligns user profiles with item personas for improved relevance.
- Offers intuitive, review-grounded explanations for recommendations.
Computer Science > Information Retrieval arXiv:2602.21756 (cs) [Submitted on 25 Feb 2026] Title:Offline Reasoning for Efficient Recommendation: LLM-Empowered Persona-Profiled Item Indexing Authors:Deogyong Kim, Junseong Lee, Jeongeun Lee, Changhoe Kim, Junguel Lee, Jungseok Lee, Dongha Lee View a PDF of the paper titled Offline Reasoning for Efficient Recommendation: LLM-Empowered Persona-Profiled Item Indexing, by Deogyong Kim and 6 other authors View PDF HTML (experimental) Abstract:Recent advances in large language models (LLMs) offer new opportunities for recommender systems by capturing the nuanced semantics of user interests and item characteristics through rich semantic understanding and contextual reasoning. In particular, LLMs have been employed as rerankers that reorder candidate items based on inferred user-item relevance. However, these approaches often require expensive online inference-time reasoning, leading to high latency that hampers real-world deployment. In this work, we introduce Persona4Rec, a recommendation framework that performs offline reasoning to construct interpretable persona representations of items, enabling lightweight and scalable real-time inference. In the offline stage, Persona4Rec leverages LLMs to reason over item reviews, inferring diverse user motivations that explain why different types of users may engage with an item; these inferred motivations are materialized as persona representations, providing multiple, human-interpretable v...
