| dc.contributor | 企管系 | |
| dc.creator (作者) | 侯佩妤 | |
| dc.creator (作者) | Ho, Yi-Mo;Ku, Wei-Pin;Hou, Pei-Yu | |
| dc.date (日期) | 2025-12 | |
| dc.date.accessioned | 9-Sep-2026 16:42:02 (UTC+8) | - |
| dc.date.available | 9-Sep-2026 16:42:02 (UTC+8) | - |
| dc.date.issued (上傳時間) | 9-Sep-2026 16:42:02 (UTC+8) | - |
| dc.identifier.uri (URI) | https://ah.lib.nccu.edu.tw/item?item_id=184875 | - |
| dc.description.abstract (摘要) | Supply chain disruptions pose significant risks to global operations, often magnified by limited data visibility and fragmented information sharing among firms. Knowledge graphs (KGs) provide a promising approach to integrate heterogeneous supply chain information into a structured, analyzable format. Traditional KG construction methods rely heavily on rule-based systems and structured data, which are often unavailable due to data privacy concerns. Recent advances in large language models (LLMs) offer new opportunities for automating KG construction; however, the use of LLM prompts struggles with domain-specific entity recognition, particularly when using lightweight LLMs, leading to reduced precision and increased noise. Moreover, their reliance on cloud-based services raises concerns regarding cost, privacy, and data security. To address these issues, we propose LiteKG, a lightweight and locally deployable framework for domain-specific KG construction. The framework is domain-agnostic by design, yet can be easily adapted to domain-specific applications. We report the results of experiments on information extraction tasks using supply chain data. Our findings show that LiteKG offers a practical and privacy-conscious alternative to build KGs in specialized domain, easing the cost and noise issues common in AI-driven methods, particularly in supply chain applications but also extendable to other domains. | |
| dc.format.extent | 114 bytes | - |
| dc.format.mimetype | text/html | - |
| dc.relation (關聯) | 2025 IEEE International Conference on Big Data (BigData), IEEE, pp.5652-5661 | |
| dc.subject (關鍵詞) | supply chain visibility; knowledge-graph construction; large language models; natural language processing | |
| dc.title (題名) | LiteKG: A Lightweight LLM-Assisted Framework for Domain-Specific Knowledge Graph Construction | |
| dc.type (資料類型) | conference | |
| dc.identifier.doi (DOI) | 10.1109/BigData66926.2025.11401070 | |
| dc.doi.uri (DOI) | https://doi.org/10.1109/BigData66926.2025.11401070 | |
