LeTS: Learning to Think-and-Search via Process-and-Outcome Reward Hybridization

TL;DR

LeTS introduces a hybrid reward framework for LLMs that enhances reasoning by integrating process-level and outcome-level rewards, improving generalization and efficiency in retrieval-augmented generation tasks.

Contribution

The paper presents a novel LeTS framework that combines process and outcome rewards to improve reasoning in LLMs without extra annotations.

Findings

LeTS improves reasoning accuracy across multiple benchmarks.

The hybrid reward approach enhances inference efficiency.

Process-and-outcome reward hybridization boosts reasoning capabilities.

Abstract

Large language models (LLMs) have demonstrated impressive capabilities in reasoning with the emergence of reasoning models like OpenAI-o1 and DeepSeek-R1. Recent research focuses on integrating reasoning capabilities into the realm of retrieval-augmented generation (RAG) via outcome-supervised reinforcement learning (RL) approaches, while the correctness of intermediate think-and-search steps is usually neglected. To address this issue, we design a process-level reward module to mitigate the unawareness of intermediate reasoning steps in outcome-level supervision without additional annotation. Grounded on this, we propose Learning to Think-and-Search (LeTS), a novel framework that hybridizes stepwise process reward and outcome-based reward to current RL methods for RAG. Extensive experiments demonstrate the generalization and inference efficiency of LeTS across various RAG benchmarks. In addition, these results reveal the potential of process- and outcome-level reward hybridization in boosting LLMs' reasoning ability via RL under other scenarios. The code will be released soon.