JCSE, vol. 20, no. 3, pp.154-165, September, 2026

DOI: http://dx.doi.org/10.5626/JCSE.2026.20.3.154

Automatic Summarization Method for Logistics Information Integrating Semantic Graph Structure and Hierarchical Attention

Jingliang Zhu
Teaching Affairs Department, Jiangsu Vocational Institute of Commerce, Nanjing, China

Abstract: In response to challenges in processing unstructured logistics text and poor domain adaptability of traditional summarization methods, this study proposes an automatic summarization approach integrating semantic graph structure and hierarchical attention. A deep semantic graph is built using a Transformer-based bidirectional encoder, combined with a message passing neural network to enhance reasoning over entities and relationships. Key nodes are selected via TextRank, and a graph-to-sequence model with hierarchical attention generates coherent summaries. Experimental results show strong performance: throughput reached 75.0 texts/min, ROUGE score increased to 0.64, and accuracy in extracting domain-specific terms from warehousing records reached 94.5%. Under 20% text noise, the F1-score was 0.58, and semantic coherence scored 4.3. The method demonstrates effectiveness in efficiency, quality, domain adaptation, robustness, and readability, supporting intelligent logistics applications.

Keyword: No keyword

Full Paper:   0 Downloads, 2 View