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

Minsang Kim,Cheoneum Park,Seungjun Baek

2024-09-27

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{<a class="link-external link-https" href="https://github.com/kmswin1/QPaug" rel="external noopener nofollow">this https URL</a>}.

Computation and Language,Artificial Intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to improve the retrieval performance and answer - generation quality of large - language models (LLMs) when dealing with complex or ambiguous questions in open - domain question - answering tasks. Specifically, although the existing retrieval - augmented generation (RAG) methods can use external knowledge sources to supplement the parametric knowledge of LLMs, when faced with complex or ambiguous questions, the retrieved paragraphs may contain irrelevant or distracting information, thus affecting the quality of the answers. Therefore, the paper proposes a simple and effective method - Question and Paragraph Augmentation (QPaug). By decomposing the original question into multiple sub - questions and using LLMs to automatically generate supplementary paragraphs to guide answer extraction, the retrieval quality and answer accuracy are improved. This method aims to improve the performance of open - domain question - answering tasks by combining the parametric knowledge and non - parametric knowledge of LLMs.