Grounding by Trying: LLMs with Reinforcement Learning-Enhanced Retrieval

TL;DR

LeReT is a reinforcement learning framework that trains LLMs to generate better search queries by trying different options and learning from successful results, significantly improving retrieval accuracy and answer quality.

Contribution

Introduces LeReT, a novel RL-based method enabling LLMs to learn effective search queries through trial and error, enhancing retrieval and grounding performance.

Findings

Up to 29% improvement in retrieval accuracy

Up to 17% enhancement in downstream answer quality

Applicable to various off-the-shelf retrievers

Abstract

The hallucinations of large language models (LLMs) are increasingly mitigated by allowing LLMs to search for information and to ground their answers in real sources. Unfortunately, LLMs often struggle with posing the right search queries, especially when dealing with complex or otherwise indirect topics. Observing that LLMs can learn to search for relevant facts by $\textit{trying}$ different queries and learning to up-weight queries that successfully produce relevant results, we introduce $\underline{Le}$arning to $\underline{Re}$trieve by $\underline{T}$rying (LeReT), a reinforcement learning framework that explores search queries and uses preference-based optimization to improve their quality. LeReT can improve the absolute retrieval accuracy by up to 29% and the downstream generator evaluations by 17%. The simplicity and flexibility of LeReT allows it to be applied to arbitrary off-the-shelf retrievers and makes it a promising technique for improving general LLM pipelines. Project website: http://sherylhsu.com/LeReT/.