基于多特征融合及奖惩机制的藏医药领域实体关系联合抽取

Joint Extraction of Entity Relations in the Field of Tibetan MedicineBased on Multi-Feature Fusion and Reward-and-Punishment Mechanism

实体关系联合抽取任务旨在识别命名实体的同时可抽取实体间的语义关系。该文提出了一种基于多特征融合及奖惩机制的藏医药领域实体关系联合抽取方法,针对基于序列标注的联合抽取方法中标注策略的局限性及特征单一、模型学习能力有限的问题,提出以下解决方案:①使用嵌套实体标注策略突破原有标注方法的局限;②使用类别特征静态融合、多特征动态融合方法及奖惩机制分别用于特征增强及模型优化。实验结果表明,该文方法提升了藏医药领域联合抽取模型的效果,模型最终的F 1值为79.23%。同时,为了证明该文模型的鲁棒性及有效性,还在SKE及NYT领域数据上进行了相关实验,实验结果验证该模型的有效性,且优于基线方法。

The entity relation joint extraction task refers to extracting semantic relations between entities while identifying named entities.This paper proposes a joint extraction method of entity relations in the Tibetan medicine field based on multi feature fusion and reward-and-punishment mechanism.We adopt the nested entity annotation strategy to break through the limitations of existing annotation methods.The static fusion of category features,dynamic fusion of multi features,and reward-and-punishment mechanisms are applied for feature enhancement and model optimization.The experimental results show that our method is effective and superior to the baseline methods.