Abstract: Abstract: To address the limitations of existing complex event extraction methods in event segmentation, particularly their inability to handle cases where the same trigger word evokes multiple events of the same type, a novel framework based on event element relation recognition and complete subgraph search was proposed. First, a concise and innovative word-pair relation labeling schema was designed, incorporating Span relations to identify event element boundaries and EI relations to determine whether elements belonged to the same event. Second, a single-stage word-pair relation recognition model was constructed, where text representations were obtained through an encoding layer, event type information was injected via an event information fusion layer, and word-pair relations were predicted using a distance-aware scoring function in the prediction layer. Finally, based on the predicted EI relations, an undirected graph was built, and a recursive complete subgraph search algorithm was designed to partition event elements, theoretically enabling the exhaustive extraction of all complex event patterns. Experimental results on the Chinese event extraction datasets FewFC and DuEE demonstrated that the proposed method significantly outperformed existing approaches in event-level F1-score, particularly achieving a 21.5% improvement in overlapping events with the same trigger. The framework effectively resolved the issue of same-type trigger overlap and ensured comprehensive extraction of complex events.

Key words: Keywords: Complex Event Extraction, Word-Pair Relation Modeling, Event Partitioning, Maximal Clique Search, Natural Language Processing

摘要： 摘 要: 针对现有复杂事件抽取方法在事件划分上存在的缺陷，尤其是无法处理同一触发词触发多个相同类型事件的问题，提出了一种基于事件要素关系识别和完全子图搜索的复杂事件抽取框架。首先，设计了一种简洁新颖的词对关系标签体系，包含Span关系用于识别事件要素的边界，EI关系用于表示事件要素之间是否属于同一事件。其次，构建了单阶段词对关系识别模型，通过编码层获取文本表示，事件信息融合层注入事件类型信息，预测层使用距离感知的打分函数识别词对关系。最后，基于预测的EI关系构建无向图，设计了一个递归的完全子图搜索算法对事件要素进行划分，理论上可以完备地抽取所有模式的复杂事件。在FewFC和DuEE两个中文事件抽取数据集上的实验表明，本文方法在事件级别抽取F1分数上显著优于现有方法，尤其在同类型触发词重叠事件上的表现提升了21.5%。该方法能有效解决同类型触发词重叠问题，完备地抽取复杂事件。

关键词: 关键词: 复杂事件抽取, 词对关系建模, 事件划分, 完全子图搜索, 自然语言处理