【摘要】 部分联合学习模型使用同一个编码器对实体识别和关系抽取这两项任务进行编码,但是单个编码器不足以捕获同一空间中两个任务所需的信息,本文通过使用序列编码器获取实体标签,表格编码器获取关系标签的双编码方式改善这一缺陷;与此同时,为了让整个模型更快、更好的收敛到理想效果,在双编码器模型的基础上加入了最小风险训练来优化全局损失函数。与现有的主流模型在4个标准数据集上进行对比发现,本文模型相较主流模型在评价指标上均有一定程度的提升。更多还原

【Abstract】 Some of the joint learning model use the same encoder to encode the two tasks of entity recognition and relationship extraction,but a single encoder is not enough to capture the information required by the two tasks in the same space. By using the sequence encoder to obtain the entity tags and the table encoder to obtain the relationship tags,the Two-Encoders can improve this defect. At the same time,in order to make the entire model converge to the desired effect faster and better,minimum risk training is added to the Two-Encoders model to optimize the global loss function. Comparing with the existing mainstream models on the four standard data sets,it is found that the model in this paper has a certain degree of improvement in evaluation indicators compared with the mainstream models.更多还原