ETC-KG: A Hybrid Extraction and Cross-Modal Alignment Method for Knowledge Graph Construction in the ETC

The Electronic Toll Collection (ETC) system is a nationwide highway tolling infrastructure that generates large-scale heterogeneous cross-system business data, which Knowledge Graphs (KGs) integrate in a structured and interpretable manner to support intelligent risk control. However, constructing high quality KGs in this domain faces significant challenges, such as complex multi-source heterogeneous data structures, inconsistent entity identifiers across platforms, and the absence of robust knowledge update mechanisms. To address these issues, this paper proposes ETC-KG, a comprehensive framework for constructing and managing KGs in the ETC domain. The framework is built upon four key technical contributions: (1) a hybrid knowledge extraction method that integrates rule-based templates with deep semantic models, achieving an F1 score of 0.93, (2) a novel lightweight cross-modal alignment mechanism, Light-CLAP, which enhances speech-text alignment accuracy by 13% through contrastive learning, (3) a dynamic knowledge management system featuring a confidence propagation model to ensure data quality and support real time updates, and (4) a graph-based risk exploration module that facilitates community detection and risk propagation analysis. Extensive experiments on real world ETC data show that our method consistently outperforms strong baselines and representative state-of-the-art methods. The resulting KG, comprising 376,218 nodes and 530,539 edges, validates the framework’s effectiveness in enabling intelligent risk analysis and practical industrial applications.