结合规则蒸馏的情感原因发现

Emotion cause extraction using rule distillation

现有基于深度学习的情感原因发现方法往往缺乏对文本子句之间关系的建模,且存在学习过程不易控制、可解释性差和对高质量标注数据依赖过大的不足。为此,该文提出了一种结合规则蒸馏和层级注意力网络的情感原因发现方法。该方法使用结合位置编码和残差结构的层级注意力网络捕获子句内部以及子句和情感表达句之间的潜层语义关系。进而,采用基于对抗学习的知识蒸馏框架将情感原因表达相关的语言学规则引入模型,最终实现结合深度神经网络和语言学规则的情感原因发现。在中文情感原因发现数据集上的实验结果显示,该方法F1值比现有最优方法提升约0.02,达到了已知的最佳性能。

Most existing deep learning emotion cause extraction methods are unable to model latent semantic relationships between clauses. In addition, these methods are not easily controlled, are difficult to interpret and need high-quality annotations. This paper presents an emotion cause extraction method that incorporates rule distillation with a hierarchical attention network. The hierarchical attention network uses position encoding and the residual structure to capture the latent semantic relationships within the clauses and between the clauses and the emotional expression. A knowledge distillation architecture based on adversarial learning then introduces linguistic rules related to the emotion cause expression into the deep neural network. Tests on a Chinese emotion cause extraction dataset show that this method outperforms the state-of-the-art method by 0.02 in F1, the best known result.