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

TL;DR

This paper introduces AuthGR, a framework that integrates authority into generative retrieval for web search, improving trustworthiness and user engagement in high-stakes domains.

Contribution

AuthGR is the first framework to incorporate authority into generative retrieval, combining multimodal authority scoring, a three-stage training pipeline, and a hybrid ensemble for deployment.

Findings

AuthGR enhances authority and accuracy in retrieval tasks.

A 3B model matches a 14B baseline in performance.

Online tests show improved user engagement and reliability.

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.