Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-view Content-aware Indexing for Long Document Retrieval

Kuicai Dong, Derrick Goh Xin Deik, Yi Quan Lee, Hao Zhang, Xiangyang Li, Cong Zhang, Yong Liu

TL;DR

The paper tackles the challenge of long-document question answering by identifying the shortcomings of fixed-length chunking. It proposes MC-indexing, which segments structured documents at content boundaries and represents each chunk in raw text, keywords, and a generated summary, enabling unsupervised, plug-and-play retrieval across existing retrievers. The authors introduce WikiWeb2M and augmented Natural Questions to evaluate long-document QA, and show that MC-indexing substantially improves recall and answer quality across multiple retrievers, especially when using the multi-view representation. Key contributions include a content-aware chunking strategy, a tri-view chunk representation, and a dataset framework for long-doc QA that preserves document structure. The approach demonstrates practical impact by enabling better retrieval for long, structured documents without requiring retraining, with potential extensions to exploit document hierarchy and richer cross-view embeddings in future work.

Abstract

Long document question answering (DocQA) aims to answer questions from long documents over 10k words. They usually contain content structures such as sections, sub-sections, and paragraph demarcations. However, the indexing methods of long documents remain under-explored, while existing systems generally employ fixed-length chunking. As they do not consider content structures, the resultant chunks can exclude vital information or include irrelevant content. Motivated by this, we propose the Multi-view Content-aware indexing (MC-indexing) for more effective long DocQA via (i) segment structured document into content chunks, and (ii) represent each content chunk in raw-text, keywords, and summary views. We highlight that MC-indexing requires neither training nor fine-tuning. Having plug-and-play capability, it can be seamlessly integrated with any retrievers to boost their performance. Besides, we propose a long DocQA dataset that includes not only question-answer pair, but also document structure and answer scope. When compared to state-of-art chunking schemes, MC-indexing has significantly increased the recall by 42.8%, 30.0%, 23.9%, and 16.3% via top k= 1.5, 3, 5, and 10 respectively. These improved scores are the average of 8 widely used retrievers (2 sparse and 6 dense) via extensive experiments.

Multi-view Content-aware Indexing for Long Document Retrieval

TL;DR

The paper tackles the challenge of long-document question answering by identifying the shortcomings of fixed-length chunking. It proposes MC-indexing, which segments structured documents at content boundaries and represents each chunk in raw text, keywords, and a generated summary, enabling unsupervised, plug-and-play retrieval across existing retrievers. The authors introduce WikiWeb2M and augmented Natural Questions to evaluate long-document QA, and show that MC-indexing substantially improves recall and answer quality across multiple retrievers, especially when using the multi-view representation. Key contributions include a content-aware chunking strategy, a tri-view chunk representation, and a dataset framework for long-doc QA that preserves document structure. The approach demonstrates practical impact by enabling better retrieval for long, structured documents without requiring retraining, with potential extensions to exploit document hierarchy and richer cross-view embeddings in future work.

Abstract

Long document question answering (DocQA) aims to answer questions from long documents over 10k words. They usually contain content structures such as sections, sub-sections, and paragraph demarcations. However, the indexing methods of long documents remain under-explored, while existing systems generally employ fixed-length chunking. As they do not consider content structures, the resultant chunks can exclude vital information or include irrelevant content. Motivated by this, we propose the Multi-view Content-aware indexing (MC-indexing) for more effective long DocQA via (i) segment structured document into content chunks, and (ii) represent each content chunk in raw-text, keywords, and summary views. We highlight that MC-indexing requires neither training nor fine-tuning. Having plug-and-play capability, it can be seamlessly integrated with any retrievers to boost their performance. Besides, we propose a long DocQA dataset that includes not only question-answer pair, but also document structure and answer scope. When compared to state-of-art chunking schemes, MC-indexing has significantly increased the recall by 42.8%, 30.0%, 23.9%, and 16.3% via top k= 1.5, 3, 5, and 10 respectively. These improved scores are the average of 8 widely used retrievers (2 sparse and 6 dense) via extensive experiments.
Paper Structure (26 sections, 11 figures, 6 tables)

This paper contains 26 sections, 11 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Retrieval Methods.
  4. Chunking Methods.
  5. Long Document Retrieval.
  6. Methodology
  7. Content-aware Chunking
  8. Multi-View Indexing and Retrieval
  9. Dataset Construction
  10. Wikipedia Webpage 2M (WikiWeb2M)
  11. Natural Questions (NQ)
  12. Experiment
  13. Baseline Systems
  14. Evaluation Metrics
  15. Main Results
  16. ...and 11 more sections

Figures (11)

  • Figure 1: Bad cases from fixed-length chunking due to relevant text missing and inclusion of irrelevant text.
  • Figure 2: Commercial LLM on Span-QA retrieval using full document vs dedicated section.
  • Figure 3: Comparison between conventional fixed length chunking and content-aware LLM QA systems.
  • Figure 4: The evaluation results of answer generation.
  • Figure 5: Pie chart of question type distribution.
  • ...and 6 more figures