$\text{R}^2\text{R}$: A Route-to-Rerank Post-Training Framework for Multi-Domain Decoder-Only Rerankers
This addresses domain specialization in high-stakes fields like finance and law, offering a modular approach to enhance reranker performance without catastrophic forgetting, though it is incremental as it builds on existing reranker frameworks.
The paper tackles the problem of domain-specific nuances in decoder-only rerankers for Retrieval-Augmented Generation, introducing R2R with Entity Abstraction for Generalization and a Latent Semantic Router to improve cross-domain robustness, achieving consistent performance gains over generalist and fine-tuned baselines across legal, medical, and financial domains.
Decoder-only rerankers are central to Retrieval-Augmented Generation (RAG). However, generalist models miss domain-specific nuances in high-stakes fields like finance and law, and naive fine-tuning causes surface-form overfitting and catastrophic forgetting. To address this challenge, we introduce R2R, a domain-aware framework that combines dynamic expert routing with a two-stage training strategy, Entity Abstraction for Generalization (EAG). EAG introduces a counter-shortcut mechanism by masking the most predictive surface cues, forcing the reranker to learn domain-invariant relevance patterns rather than memorizing dataset-specific entities. To efficiently activate domain experts, R2R employs a lightweight Latent Semantic Router that probes internal representations from the frozen backbone decoder to select the optimal LoRA expert per query. Extensive experiments across different reranker backbones and diverse domains (legal, medical, and financial) demonstrate that R2R consistently surpasses generalist and single-domain fine-tuned baselines. Our results confirm that R2R is a model-agnostic and modular approach to domain specialization with strong cross-domain robustness.