<p>Event extraction aims to identify event triggers and their associated arguments from unstructured text. Chinese event extraction is particularly challenging due to the complex sentence structure, ambiguous word boundaries, and the nested and overlapped nature of event parameters. Previous methods have been limited by character-based approaches and reliance on external lexical information, which often lacks contextual relevance, potentially introducing noise and negatively impacting model performance. To overcome these difficulties, we have organized and annotated a fine-grained event dataset related to governance texts from real-world Chinese event data, covering flat, overlapped, and nested events. Subsequently, we propose HiEE, a novel Chinese event extraction model that combines hierarchical feature fusion and external semantic enhancement. We reformulate the task as trigger-argument relationship classification, addressing the complexities of nested and discontinuous event parameters. The Hiformer module integrates character, lexical, and syntactic embeddings, while minimizing the interference of irrelevant external information on the original text embeddings. Moreover, through the design of a dual-path gated fusion mechanism, we effectively combine linguistic features, enhancing trigger-argument representation and capturing multi-level semantics. Extensive experiments conducted on four benchmark datasets demonstrate that the proposed HiEE model achieves state-of-the-art performance, outperforming previous best baseline models. On average, HiEE yields an F1 score improvement of 1.18% across all datasets, highlighting the significant benefits brought by hierarchical feature fusion and external semantic enhancement for Chinese event extraction tasks.</p>