SLIDE: Sliding Localized Information for Document Extraction

2025-03-23

AI Summary

To address performance degradation in entity/relation extraction for knowledge graph construction under long-text and low-resource language settings—primarily caused by limited long-range modeling capacity of large language models and critical information truncation due to scarce training data—this paper proposes a sliding localization chunking mechanism. It employs overlapping sliding windows to generate semantically coherent local contexts, systematically ensuring cross-window semantic consistency and preservation of key contextual information for the first time. Furthermore, it integrates local-context enhancement, a GraphRAG-based framework, and multilingual adaptation—including Afrikaans. Experiments demonstrate substantial improvements: F1 scores for English entity/relation extraction increase by 24% and 39%, respectively, while Afrikaans achieves gains of 49% and 82%. In question-answering, comprehensive metrics—including comprehensiveness, diversity, and empowerment—attain state-of-the-art performance.

Technology Category

• Natural Language Processing: Information Extraction
• Data Mining & Knowledge Management: Linked Open Data, Knowledge Graphs & KB Completion
• Machine Learning: Graph-based Machine Learning

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Constructing accurate knowledge graphs from long texts and low-resource languages is challenging, as large language models (LLMs) experience degraded performance with longer input chunks. This problem is amplified in low-resource settings where data scarcity hinders accurate entity and relationship extraction. Contextual retrieval methods, while improving retrieval accuracy, struggle with long documents. They truncate critical information in texts exceeding maximum context lengths of LLMs, significantly limiting knowledge graph construction. We introduce SLIDE (Sliding Localized Information for Document Extraction), a chunking method that processes long documents by generating local context through overlapping windows. SLIDE ensures that essential contextual information is retained, enhancing knowledge graph extraction from documents exceeding LLM context limits. It significantly improves GraphRAG performance, achieving a 24% increase in entity extraction and a 39% improvement in relationship extraction for English. For Afrikaans, a low-resource language, SLIDE achieves a 49% increase in entity extraction and an 82% improvement in relationship extraction. Furthermore, it improves upon state-of-the-art in question-answering metrics such as comprehensiveness, diversity and empowerment, demonstrating its effectiveness in multilingual and resource-constrained settings.

Problem

• Improving knowledge graph construction from long texts
• Enhancing entity and relationship extraction in low-resource languages
• Addressing information loss in long documents for LLMs

Innovation

• Overlapping windows for local context retention
• Enhanced entity and relationship extraction accuracy
• Improved multilingual and low-resource performance