From Relevance to Authority: Authority-aware Generative Retrieval in Web Search Engines

2026-04-15

AI Summary

This work addresses the critical limitation of existing generative retrieval methods, which prioritize relevance while neglecting document authority—leading to the propagation of unreliable information in high-stakes domains such as healthcare and finance. To bridge this gap, we propose AuthGR, a novel authority-aware generative retrieval framework that, for the first time, integrates authority modeling into generative retrieval. AuthGR leverages vision-language models to fuse multimodal signals for authority scoring and employs a three-stage progressive training strategy alongside a hybrid ensemble inference mechanism. Experimental results demonstrate that a 3B-parameter AuthGR model matches the offline performance of a 14B baseline, while large-scale online A/B tests and human evaluations confirm its significant improvements in user engagement and result reliability.

Technology Category

• Natural Language Processing: Generation
• Computer Vision: Image and Video Retrieval
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Generative information retrieval (GenIR) formulates the retrieval process as a text-to-text generation task, leveraging the vast knowledge of large language models. However, existing works primarily optimize for relevance while often overlooking document trustworthiness. This is critical in high-stakes domains like healthcare and finance, where relying solely on semantic relevance risks retrieving unreliable information. To address this, we propose an Authority-aware Generative Retriever (AuthGR), the first framework that incorporates authority into GenIR. AuthGR consists of three key components: (i) Multimodal Authority Scoring, which employs a vision-language model to quantify authority from textual and visual cues; (ii) a Three-stage Training Pipeline to progressively instill authority awareness into the retriever; and (iii) a Hybrid Ensemble Pipeline for robust deployment. Offline evaluations demonstrate that AuthGR successfully enhances both authority and accuracy, with our 3B model matching a 14B baseline. Crucially, large-scale online A/B tests and human evaluations conducted on the commercial web search platform confirm significant improvements in real-world user engagement and reliability.

Problem

• generative information retrieval
• authority
• trustworthiness
• web search
• relevance

Innovation

• Authority-aware Retrieval
• Generative Information Retrieval
• Multimodal Authority Scoring
• Three-stage Training Pipeline
• Hybrid Ensemble Pipeline