Hierarchical Entity Disambiguation for Aviation Maintenance Knowledge Graph Construction: A Multi-Relational CRF Approach
Year: 2026
Type: article
Abstract: This paper presents a systematic approach for building the entity-disambiguation core of aviation maintenance knowledge graphs (KGs) from unstructured text, addressing the key challenge of cross-document entity disambiguation. We develop a three-stage pipeline: 1) domain-adapted bidirectional long short-term memory with conditional random fields (BiLSTM-CRF) for named entity recognition (NER) over a core five-type aviation ontology; 2) Canopy-based indexing for tractable candidate reduction prior to collective inference; and 3) multi-relational CRF with graph-based disambiguation exploiting hierarchical relationships between aircraft, systems, and components. The empirical evaluation in this study focuses on the two foundation tasks that feed KG construction—NER and entity linking. Evaluation on a fixed 73-document aviation benchmark yields 89.6% exact-span F1 for NER. On the 209 labeled entity-linking instances, the proposed multi-relational CRF achieves 69.1% precision, 58.9% recall, and 63.6% F1, remaining competitive with strong lexical baselines while preserving structurally informed inference. A companion reproducibility package provides the data splits, evaluation scripts, quantitative artifacts, and figure-generation workflow used in this study.
Source: IEEE Access
Authors: Wei Huang, Tianran Ma
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,075
OpenAlex ID: W7221799719
