DecoupleSearch: Decouple Planning and Search via Hierarchical Reward Modeling

TL;DR

DecoupleSearch introduces a framework that separates planning and search in hierarchical reasoning, improving the efficiency and effectiveness of retrieval-augmented generation systems with dual value models and Monte Carlo Tree Search.

Contribution

It proposes a novel decoupling approach for planning and search in RAG systems using dual value models and hierarchical beam search, enabling independent optimization.

Findings

Effective across various policy model sizes

Improves reasoning tree construction and search grounding

Enhances flexibility and performance of RAG systems

Abstract

Retrieval-Augmented Generation (RAG) systems have emerged as a pivotal methodology for enhancing Large Language Models (LLMs) through the dynamic integration of external knowledge. To further improve RAG's flexibility, Agentic RAG introduces autonomous agents into the workflow. However, Agentic RAG faces several challenges: (1) the success of each step depends on both high-quality planning and accurate search, (2) the lack of supervision for intermediate reasoning steps, and (3) the exponentially large candidate space for planning and searching. To address these challenges, we propose DecoupleSearch, a novel framework that decouples planning and search processes using dual value models, enabling independent optimization of plan reasoning and search grounding. Our approach constructs a reasoning tree, where each node represents planning and search steps. We leverage Monte Carlo Tree Search to assess the quality of each step. During inference, Hierarchical Beam Search iteratively refines planning and search candidates with dual value models. Extensive experiments across policy models of varying parameter sizes demonstrate the effectiveness of our method.