融合对抗学习的因果关系抽取

因果关系抽取在事件预测、情景生成、问答以及文本蕴涵等任务上都有重要的应用价值.但多数现有的因果关系抽取方法都需要人工定义模式和约束，且严重依赖知识库.为此，本文利用生成式对抗网络（Generative adversarial networks，GAN）的对抗学习特性，将带注意力机制的双向门控循环单元神经网络（Bidirectional gated recurrent units networks，BGRU）与对抗学习相融合，通过重定义生成模型和判别模型，基本的因果关系抽取网络能够与判别网络形成对抗，进而从因果关系解释信息中获得高区分度的特征.实验结果表明，与当前用于因果关系抽取的方法相比较，该方法表现出更优的抽取效果.

Causality Extraction With GAN

Causality extraction is of important practical value in tasks such as event prediction, scenario generation, question answering, and textual implication; but most of the existing causality extraction methods require artificial definition of patterns and constraints and are heavily dependent on knowledge base. In this paper, the bidirectional gated recurrent units networks (BGRU) with attention mechanism are merged with confrontational learning by leveraging the confrontational learning characteristics of generative adversarial networks (GAN). Through redefining the generator and discriminator, the basic causality extraction network can construct a confrontation with the discriminator, and then obtain a high distinguishing feature from the causality interpretation information. Our experiments show that our approach leads to an improved performance over strong baselines.