GraphLM: A Comprehensive Survey on Knowledge Graphs for Intelligent Document Understanding

Abstract

The unstructured growth on the textual data in the academic and industrial sectors poses a serious challenge in knowledge extraction and understanding relationships. This survey explores the knowledge graph construction, retrieval- augmented generation models and how these models can be integrated to have document comprehension systems. By analytically examining nineteen methods such as ChatGPT, Semantic Scholar, and specialized systems, we determine serious deficiencies in integrating facts in a unified format, mitigating hallucinations, and being able to explain. GraphLM is a suggestion of a platform that combines knowledge graphs with RAG to provide structured and verifiable insights. Available systems have hallucination rates of 28-39% and can be used in only 15% of the cases of necessary visualization. GraphLM has 50-70% reduced hallucinations, 92- 95% entity extraction precision and 100% claim traceability which are provided with confidence-weighted knowledge graphs, semantic entity extraction pipelines, and interactive visualization structures.

Introduction

Unstructured text data is growing rapidly, creating challenges in computational efficiency, pattern extraction, and factual reliability. Traditional information retrieval systems are limited in understanding complex semantic relationships, while large language models (LLMs) excel at natural language tasks but suffer from hallucinations—factually incorrect but plausible outputs. Knowledge graphs, which represent entities and relationships in structured, machine-readable formats, offer strong capabilities for data integration, reasoning, and contextual insights but face issues in scalability, accessibility, and integration with conversational AI.

This survey explores the integration of retrieval-augmented generation (RAG) with knowledge graphs to build intelligent document understanding systems. It reviews 19 platforms, identifies gaps, and proposes GraphLM, which combines LLMs’ language understanding with knowledge graphs’ structural accuracy. The framework addresses user needs (Builders, Analysts, Consumers), supports explainable AI, interactive visualization, and confidence-scored knowledge representation, aiming to improve comprehension, reduce hallucinations, and enhance fact-based knowledge extraction.

Key concepts include: knowledge graph construction, entity/relation extraction, graph databases (Neo4j, RDF Stores, hybrid solutions), and RAG pipelines that retrieve relevant context from document collections to improve LLM outputs. Literature highlights practical implementations, challenges in visualization, large-scale graph handling, and multi-hop reasoning improvements when combining RAG with graph structures. The study emphasizes the urgent need for reliable, scalable, and user-centered tools to manage unstructured data in research and industry.

Conclusion

This survey study includes methodologies of knowledge graph construction and retrieval-augmented generation of doc- ument understanding applications. After an intensive examination of nineteen currently existing solutions comprising conversational interfaces, academic search systems, and purposeful graph platforms, we find significant weaknesses in unified fact representation, hallucination mitigation, and explainability of the system. We have found that current systems have 28-39% rates of hallucinating and can only sustain 15% of the necessitated visualization applications. Graph LM fills in on these basic shortcomings with confidence-scored knowledge graphs that allow reliability evaluation, semantic entity extraction pipelines, which are up to 92-95% accurate, and interactive visualization systems, which cater to a wide range of user requirements. The proposed system is based on synergistic technologies to form superior capabilities that are more than the sum of their parts forming realistic avenues in the transformation of raw information into viable knowledge via organized representation and intelligent retrieval systems.

References

[1] H. Li, G. Appleby et al., “Knowledge Graphs in Practice: Character- izing their Users, Challenges, and Visualization Opportunities,” IEEE Transactions on Visualization and Computer Graphics, vol. 30, no. 1, pp. 584-594, 2024. [2] R. Huang et al.,Neo4j Inc., “Neo4j Graph Database Platform Documen- tation,” Neo4j Technical Report, 2024. [3] D. Edge et al., “From Local to Global: A Graph RAG Approach to Query-Focused Summarization,” arXiv preprint arXiv:2404.16130, 2024 [4] S. Ray, “ChatGPT: A comprehensive review on background, applica- tions, key challenges, bias, ethics, limitations and future scope,” Internet of Things and Cyber-Physical Systems, vol. 3, pp. 121-154, 2023. [5] J. Zhang et al., “Siren’s Song in the AI Ocean: A Survey on Halluci- nation in Large Language Models,” arXiv preprint arXiv:2309.01219, 2023. [6] A. Hogan et al., “Knowledge Graphs,” ACM Computing Surveys, vol. 54, no. 4, pp. 1-37, 2021. [7] J. Johnson, M. Douze, and H. Je´gou, “Billion-scale similarity search with GPUs,” IEEE Transactions on Big Data, vol. 7, no. 3, pp. 535- 547, 2021. [8] X. Xu et al., “LayoutLMv2: Multi-modal Pre-training for Visually- rich Document Understanding,” in Proc. 59th Annual Meeting of the Association for Computational Linguistics, 2021, pp. 2579-2591. [9] P. Lewis et al., “Retrieval-Augmented Generation for Knowledge- Intensive NLP Tasks,” in Advances in Neural Information Processing Systems, vol. 33, 2020, pp. 9459-9474. [10] A. B. Arrieta et al., “Explainable Artificial Intelligence (XAI): Con- cepts, taxonomies, opportunities and challenges toward responsible AI,” Information Fusion, vol. 58, pp. 82-115, 2020. [11] C. Miao et al.,“Calibrating Knowledge Extraction: A Case Study in Relation Extraction,” in Proc. 2020 Conf. on Empirical Methods in Natural Language Processing, 2020, pp. 3151-3161. [12] A. Akbik et al., “FLAIR: An Easy-to-Use Framework for State-of- the-Art NLP,” in Proc. 2019 Conf. North American Chapter of the Association for Computational Linguistics, 2019, pp. 54-59. [13] C. Rasmussen et al., “Schema Evolution in NoSQL Databases: A Systematic Review,” in Proc. 2019 IEEE Int. Conf. Big Data, 2019, pp. 2842-2851. [14] David Reinsel et al, “The Digitization of the World: From Edge to Core,” IDC White Paper, Doc. US44413318, 2018. [15] R. Angles et al., “Foundations of Modern Query Languages for Graph Databases,” ACM Computing Surveys, vol. 50, no. 5, pp. 1-40, 2017. [16] T. H. Nguyen et al., “Relation Extraction: Perspective from Convolu- tional Neural Networks,” in Proc. NAACL Workshop on Vector Space Modeling for NLP, 2015, pp. 39-48. [17] N. J. Van Eck and L. Waltman, “Software survey: VOSviewer, a computer program for bibliometric mapping,” Scientometrics, vol. 84, no. 2, pp. 523-538, 2010. [18] E. Curry et al., “The Role of Community-Driven Data Curation for Enterprises,” in Linking Enterprise Data, D. Wood, Ed., Springer, 2010, pp. 25-47. [19] C. Manning et al., “Introduction to Information Retrieval,” Cambridge University Press, 2008. [20] B. Lee et al., “Task taxonomy for graph visualization,” in Proc. AVI Workshop on BEyond time and errors: novel evaLuation methods for Information Visualization, 2006, pp. 1-5.

Copyright

Copyright © 2025 Ms. R. R. Owhal, Tabish Ali Ansari, Vaishnavi Pangare, Sushant Shinde. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.