MemSum-DQA: Adapting An Efficient Long Document Extractive Summarizer for Document Question Answering

TL;DR

MemSum-DQA is an efficient extractive summarization-based system for document question answering that improves accuracy by 9% over previous methods, especially excelling in child-relationship questions.

Contribution

It adapts the MemSum summarizer for DQA by prefixing questions to document blocks, achieving significant accuracy improvements and demonstrating effectiveness in complex relational questions.

Findings

9% improvement in exact match accuracy

Effective in child-relationship understanding questions

Leverages extractive summarization for DQA

Abstract

We introduce MemSum-DQA, an efficient system for document question answering (DQA) that leverages MemSum, a long document extractive summarizer. By prefixing each text block in the parsed document with the provided question and question type, MemSum-DQA selectively extracts text blocks as answers from documents. On full-document answering tasks, this approach yields a 9% improvement in exact match accuracy over prior state-of-the-art baselines. Notably, MemSum-DQA excels in addressing questions related to child-relationship understanding, underscoring the potential of extractive summarization techniques for DQA tasks.