基于文献的生物医学事件抽取是生物医学自然语言处理领域的重要研究课题，为新药的研发和疾病的辅助诊断、预防、治疗提供启发和依据。目前的方法存在数据表示语义信息不足、复杂事件抽取精度较低、级联错误和冗余信息等亟待解决的关键问题。由此本项目提出：(1)融入领域知识获取领域扩展词特征，为生物事件抽取研究构建全新的多信息数据向量表示，并将领域知识语义与文本语义有机结合，改善数据语义信息不足的问题；(2)根据简单事件和复杂事件的要素结构特点分别构建针对性的要素检测模型，同时设计多级注意力机制加强要素之间的相互作用，进一步提升复杂事件抽取性能；(3)提出新的基于混合神经网络和动态路径规划策略的联合事件抽取方法，减少冗余信息，避免分阶段方法的级联错误，最终获得高性能的生物事件抽取模型。本项目以癌症相关的生物医学文献为主要研究对象，通过构建生物医学事件数据库和交互网络，为相关疾病的研究提供有力支持。

Biomedical event extraction from literature is an important research topic in the field of biomedical natural language processing, which provides inspirations and basis for the research and development of new drugs as well as the auxiliary diagnosis, prevention and treatment of diseases. By far, there are some key issues needed to be solved urgently: insufficient semantic information of data representation, lower performance of complex event extraction, cascading errors and redundant information. Therefore, this project proposes: (1) a new multi-information data vector representation for biomedical event extraction research by integrating features of domain-extended words obtained from domain knowledge and combined domain knowledge semantics with text semantics, which improves the insufficiency of data semantic information; (2) a new targeted argument detection model designed according to the structural features of simple events and complex events, which integrates a multi-level attention mechanism, aiming to further improve the performance of complex biomedical event extraction; (3) a novel joint event extraction method based on hybrid neural network and dynamic path strategy to reduce redundant information and avoid the cascading errors in pipeline approach, so a high performance biomedical event extraction model will be obtained. This project takes cancer-related biomedical literature as the main research object, and provides strong support for related research by constructing biomedical event databases and an interactive network.