acl2021专题

当GNN遇见NLP(十一):Directed Acyclic Graph Network for Conversational Emotion Recognition,ACL2021

同上一篇一样,是使用图建模对话情绪识别的论文。本文提出了用有向无环图(DAG)对对话进行编码的方法(DAG-ERC),主要借鉴了ICLR2021发表的工作DAGNN(Directed acyclic graph neural networks.)中的GNN改进策略。DAG-ERC在DAGNN的基础上进行了两个改进:(1)基于说话人身份的关系感知特征转换,以收集信息;(2)上下文信息单元,以增强历

ACL2021_Learning from Miscellaneous Other-Class Words for Few-shot Named Entity Recognition

Learning from Miscellaneous Other-Class Words for Few-shot Named Entity Recognition 知识准备摘要介绍相关工作原型网络 方法符号标记Undefined Classes DetectionStep 1: Mapping Function LearningStep 2: Binary Group Classifie