生物医学事件触发词识别研究

Research on identification of biomedical event trigger

触发词的识别是生物医学事件抽取的一个关键步骤。传统的采用字典或规则的方法过于依赖字典或规则的建立,一般的机器学习方法则需设计复杂的特征,而且大多数系统采用串行方法会导致错误的传播。从算法和整体流程两个维度进行优化,采用了基于神经网络的事件触发词识别和事件类型判别联合结构预测模型,既简化人工干预,又减少错误传播。实验结果表明提出的方法取得了很好的性能,为生物事件抽取奠定了可靠的基础。

The trigger identification is a key step in the biomedical event extraction. The traditional method based on dictionary or rules are too dependent on the establishment of a dictionary or rules,while general machine learning methods need to design complex features,even most systems that adopt the serial method lead to the error propagation. From two dimensions of algorithm and the whole process to optimize,the system used a joint structure prediction model of event trigger identification and event type determination based on neural network. It simplified the manual intervention,and reduced error propagation. The evaluation shows that this proposed approach achieves good performance,which lays the reliable foundation of the biomedical event extraction.