Unifying Ranking and Generation in Query Auto-Completion via Retrieval-Augmented Generation and Multi-Objective Alignment

2026-02-18 · Source: Apple Machine Learning Research · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, quick

Summary

A new unified framework addresses challenges in Query Auto-Completion (QAC) by reformulating it as end-to-end list generation. This approach integrates Retrieval-Augmented Generation (RAG) and multi-objective Direct Preference Optimization (DPO) to overcome limitations of traditional retrieve-and-rank pipelines, which struggle with long-tail coverage, and generative methods, which risk hallucination. Key innovations include multi-objective optimization for list generation, a comprehensive methodology combining RAG with learned and rule-based verifiers for synthetic data, and a hybrid serving architecture for efficient production deployment. Evaluated on a large-scale commercial search platform, the framework achieved significant improvements, including +0.40 to +0.69 preference scores in human evaluation, a 5.44% reduction in keystrokes, and a 3.46% increase in suggestion adoption in online experiments.

Key takeaway

For AI Engineers developing search and recommendation systems, this framework offers a production-validated method to enhance QAC. You should consider adopting a RAG-powered, end-to-end generative approach with multi-objective DPO to improve suggestion quality and user engagement, potentially reducing keystrokes by 5.44% and increasing adoption by 3.46% in your own systems.

Key insights

QAC can be effectively reframed as end-to-end list generation using RAG and multi-objective DPO.

Principles

• Combine RAG with multi-objective DPO for QAC.
• Use learned and rule-based verifiers for data quality.
• Employ iterative critique-revision for synthetic data.

Method

Reformulate QAC as end-to-end list generation, integrating RAG, multi-objective DPO, and verifiers. Utilize iterative critique-revision for high-quality synthetic data generation, and deploy with a hybrid serving architecture.

In practice

• Apply RAG to improve long-tail QAC coverage.
• Implement DPO for multi-objective QAC alignment.
• Design hybrid architectures for low-latency QAC.

Topics

• Query Auto-Completion
• Retrieval-Augmented Generation
• Direct Preference Optimization
• Multi-objective Optimization
• End-to-End Generation

Best for: AI Engineer, NLP Engineer, AI Scientist, AI Researcher, Machine Learning Engineer, Data Scientist

Related on AIssential

• Modernizing the Facebook Groups Search to Unlock the Power of Community Knowledge — Hybrid retrieval, multi-objective ranking, and AI-driven evaluation enhance community content discovery and validation.
• Predictive Prefetching for Retrieval-Augmented Generation — Predictive prefetching in RAG significantly reduces latency by anticipating information needs before critical uncertainty.
• DocQAC: Adaptive Trie-Guided Decoding for Effective In-Document Query Auto-Completion — Adaptive trie-guided decoding improves in-document query auto-completion by leveraging document context and balancing model confidence with trie guidance.
• Vespa AI and Surpassing the Limits of Vector Search — Vector similarity alone is insufficient for robust search relevance; tensor-based retrieval offers superior flexibility and efficiency.
• An Efficient Approach for Answering Not Readily Attainable Questions for RAG-based Applications — This paper proposes an efficient approach for RAG systems to answer questions not readily found in retrieved data.
• Understanding Retrieval Augmented Generation (RAG): End-to-End Explained — RAG combines LLM reasoning with real-time data retrieval to produce accurate, context-aware, and trustworthy AI responses.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Apple Machine Learning Research.