@inproceedings{niu-etal-2025-tree,
    title = "Tree-{KG}: An Expandable Knowledge Graph Construction Framework for Knowledge-intensive Domains",
    author = "Niu, Songjie  and
      Yang, Kaisen  and
      Zhao, Rui  and
      Liu, Yichao  and
      Li, Zonglin  and
      Wang, Hongning  and
      Chen, Wenguang",
    editor = "Che, Wanxiang  and
      Nabende, Joyce  and
      Shutova, Ekaterina  and
      Pilehvar, Mohammad Taher",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingestion-acl-25/2025.acl-long.907/",
    pages = "18516--18529",
    ISBN = "979-8-89176-251-0",
    abstract = "In knowledge-intensive domains like scientific research, effective decisions rely on organizing and retrieving intricate data. Knowledge graphs (KGs) help by structuring entities, relations, and contextual dependencies, but building KGs in such domains is challenging due to inherent complexity, manual effort, and rapid evolution. Inspired by how humans organize knowledge hierarchically, we propose Tree-KG, an expandable framework that combines structured domain texts with advanced semantic techniques. First, Tree-KG builds a tree-like graph from textbook structures using large language models (LLMs) and domain-specific entities, creating an \textit{explicit KG}. Then, through iterative expansion with flexible, predefined operators, it uncovers \textit{hidden KG} while preserving semantic coherence. Experiments demonstrate that Tree-KG consistently surpasses competing methods, achieving the highest F1 scores (12{--}16{\%} above the second-best), with notable performance (F1 0.81) on the Text-Annotated dataset, highlighting its effectiveness in extracting high-quality information from source texts. Additionally, Tree-KG provides superior structural alignment, domain-specific extraction, and cost-efficiency, delivering robust results with reduced token usage and adaptable, resource-conscious deployment."
}