R3A: Reinforced Reasoning for Relevance Assessment for RAG in User-Generated Content Platforms

TL;DR

R3A introduces a reinforced reasoning approach that improves relevance assessment in RAG systems on user-generated content platforms by decomposing intent inference and evidence grounding.

Contribution

The paper presents R3A, a novel model that enhances relevance assessment by leveraging auxiliary data and explicit evidence grounding, addressing UGC-specific challenges.

Findings

R3A outperforms strong baselines on offline benchmarks.

R3A-1.5B model achieves significant online A/B testing gains.

Explicit evidence grounding reduces noise sensitivity.

Abstract

Retrieval-augmented generation (RAG) plays a critical role in user-generated content (UGC) platforms, but its effectiveness critically depends on accurate query-document relevance assessment. Despite recent advances in applying large language models (LLMs) to relevance modeling, UGC platforms present unique challenges: 1) ambiguous user intent due to sparse user feedback in RAG scenarios, and 2) asymmetric relevance, where relevance is driven by localized answer-bearing content rather than global query-document similarity. To address these issues, we propose the Reinforced Reasoning model for Relevance Assessment (R3A), which decomposes relevance assessment into intent inference and evidence grounding. R3A leverages auxiliary high-clicked documents to infer latent query intent, and extracts verbatim evidence fragments to ground relevance decisions, reducing noise sensitivity and improving asymmetric relevance modeling. Experimental results demonstrate that R3A substantially outperforms strong baselines on offline benchmarks, while the distilled R3A-1.5B model achieves significant gains in large-scale online A/B testing, effectively balancing performance and practical deployability.