Bi-NAS framework boosts recommender explanations with LLM integration
New bi-level search optimizes cross-attention and feature functions for personalized justifications.
Computer scientists from Amazon and academic institutions have introduced Bi-NAS (Bi-level Neural Architecture Search), a novel framework that optimizes explanation generation in recommender systems. Presented on arXiv on July 1, 2026, Bi-NAS simultaneously refines cross-attention mechanisms and feature interaction functions by exploring both intra-layer and inter-layer design spaces. This dual-level search allows the system to automatically discover the most effective architecture for producing explanations that are not only accurate but also personalized.
The framework integrates Large Language Models (LLMs) via zero-shot prompting to generate justifications that align user feature preferences with item quality scores. By considering both user intent and item attributes, Bi-NAS enhances transparency and reasoning depth in recommendations. Extensive tests on four real-world datasets show that the approach boosts recommendation accuracy and significantly improves explanation effectiveness, giving users clearer insights into why specific items are suggested. This work addresses a critical gap in recommender systems, where explanations are often generic or absent.
- Bi-NAS uses bi-level neural architecture search to jointly optimize cross-attention and feature interaction functions for explanations.
- LLMs with zero-shot prompting generate personalized justifications that reflect user preferences and item quality.
- Evaluated on four real-world datasets, Bi-NAS improves both recommendation accuracy and explanation effectiveness.
Why It Matters
Personalized, transparent explanations from recommenders can boost user trust and decision-making across e-commerce, media, and search platforms.