QPaug: Question and Passage Augmentation for Open-Domain Question Answering of LLMs

TL;DR

QPaug enhances open-domain QA by decomposing questions into sub-questions and augmenting retrieved passages with self-generated content, leading to improved retrieval and answer accuracy in LLM-based systems.

Contribution

Introduces QPaug, a novel method that decomposes questions and augments passages using LLMs, significantly improving retrieval and answer quality in open-domain QA.

Findings

QPaug outperforms previous state-of-the-art methods.

Significant performance gains over existing RAG approaches.

Effective handling of ambiguous or complex questions.

Abstract

Retrieval-augmented generation (RAG) has received much attention for Open-domain question-answering (ODQA) tasks as a means to compensate for the parametric knowledge of large language models (LLMs). While previous approaches focused on processing retrieved passages to remove irrelevant context, they still rely heavily on the quality of retrieved passages which can degrade if the question is ambiguous or complex. In this paper, we propose a simple yet efficient method called question and passage augmentation (QPaug) via LLMs for open-domain QA. QPaug first decomposes the original questions into multiple-step sub-questions. By augmenting the original question with detailed sub-questions and planning, we are able to make the query more specific on what needs to be retrieved, improving the retrieval performance. In addition, to compensate for the case where the retrieved passages contain distracting information or divided opinions, we augment the retrieved passages with self-generated passages by LLMs to guide the answer extraction. Experimental results show that QPaug outperforms the previous state-of-the-art and achieves significant performance gain over existing RAG methods. The source code is available at \url{https://github.com/kmswin1/QPaug}.