一种在矩阵空间中识别文本蕴涵的动态交互网络

Dynamic interactive network over matrix-space for recognizing textual entailment

针对文本蕴涵问题提出一种动态交互网络(dynamic interactive network,DIN)进行识别。不同于已有交互模型,DIN将两句词向量投射到二维矩阵空间中进行交互,然后利用输出矩阵为同时处理上下文信息和控制信息流动的GRU编码器生成动态权重。前者通过更高阶形式的信息交互挖掘深层逻辑片段,后者通过改变交互信息与上下文信息的结合模式帮助编码器有效区分两者的重要性差异。模型在SNLI测试集上获得了88.0%的识别准确度,超过已有的最佳模型,且使用的训练参数仅为它的一半。

This paper presented a DIN for recognizing textual entailment.Unlike the other interactive models,DIN facilitated the interaction by projecting the embedding vectors into a two-dimensional matrix space,and then used the output matrices to produce dynamic weights for the GRU encoder that both processed the context information and controlled the information flow.It empowered the extraction of logic segments through higher-orders of information interactions and helped the encoder better choose between the context and the interactive information.Experiments on the SNLI corpus show that the proposed model achieves a test accuracy of 88.0%,outperforming the state-of-the-art with only a small amount of the training parameters introduced.