Evaluate-as-Action: Self-Evaluated Process Rewards for Retrieval-Augmented Agents
This addresses reliability issues in multi-step reasoning for retrieval-augmented agents, offering a novel method to improve accuracy in open-domain question answering, though it is incremental in building on existing retrieval-augmented approaches.
The paper tackled the problem of unreliable multi-step reasoning in retrieval-augmented agents by proposing Evaluate-as-Action (EvalAct), which converts retrieval quality assessment into an explicit action and uses Process-Calibrated Advantage Rescaling (PCAR) for optimization, achieving the best average accuracy on seven open-domain QA benchmarks with the largest gains on multi-hop tasks.
Retrieval-augmented agents can query external evidence, yet their reliability in multi-step reasoning remains limited: noisy retrieval may derail multi-hop question answering, while outcome-only reinforcement learning provides credit signals that are too coarse to optimize intermediate steps. We propose \textsc{EvalAct} (Evaluate-as-Action), which converts implicit retrieval quality assessment into an explicit action and enforces a coupled Search-to-Evaluate protocol so that each retrieval is immediately followed by a structured evaluation score, yielding process signals aligned with the interaction trajectory. To leverage these signals, we introduce Process-Calibrated Advantage Rescaling (PCAR), a GRPO-based optimization method that rescales advantages at the segment level according to evaluation scores, emphasizing reliable segments while updating uncertain ones conservatively. Experiments on seven open-domain QA benchmarks show that \textsc{EvalAct} achieves the best average accuracy, with the largest gains on multi-hop tasks, and ablations verify that the explicit evaluation loop drives the primary improvements while PCAR provides consistent additional benefits.