A BERT-Based Deep Model for Contract Element Extraction with Multi-Scale Attention
This conference paper proposes a BERT-based deep learning model with multi-scale attention for contract element extraction in legal text analysis. The work addresses limitations of keyword matching and template-based approaches by modeling contextual information, clause dependencies, and legal expression structures to improve the accuracy and practical usability of contract element extraction.
Presented as a poster at the 2026 4th IEEE International Conference on Pattern Recognition, Machine Vision and Artificial Intelligence (PRMVAI 2026). Authors visible on the poster include Ning Mou, Shuo Han, Xiu-Fu Ye, and Fan Wu.
基於BERT並結合多尺度注意力的合約要素抽取深度模型
本研究面向合約文本分析中的合約要素抽取任務，提出一種以 BERT 為基礎、結合多尺度注意力機制的深度學習模型。針對傳統關鍵詞比對和模板比對方法難以處理複雜條款、隱藏責任和長距離依賴的問題，該模型透過捕捉上下文資訊、條款間依賴關係與法律表達結構，提升合約要素抽取的準確率和實用性。
該論文於2026年第四屆IEEE模式識別、機器視覺與人工智慧國際會議（PRMVAI 2026）以海報形式發表。海報列出的作者包括 Ning Mou、Shuo Han、Xiu-Fu Ye、Fan Wu。