融合分段编码与仿射机制的相似案例匹配方法

相似案例匹配任务旨在判断2篇裁判文书所描述的案件是否相似,通常被看作裁判文书的文本匹配问题,在司法审判过程中具有重要的应用。现有深度学习模型大多将案例长文本编码为单一向量表示,模型很难从长文本中学习到裁判文书之间的细微差异。考虑到案例文本各部分的内容较为固定,本文提出将案例长文本拆分为多个片断并分别编码,以便获取不同部分的细微特征;同时,采用可学习仿射变换改进相似度打分模块,使模型学习到了更多细微的差异,进一步提高了案例匹配的性能。在CAIL2019-SCM数据集上的实验结果表明,本文提出方法与现有方法相比准确率提升了1.89%。

A similarity case matching method combining segment encoding and affine-mechanism

Similarity case matching(SCM)task is to judge whether the cases described in two judgment documents are similar. SCM is usually regarded as the text matching problem of judgment documents and has important applications in the judicial trial. Existing deep learning models mostly encode long texts of cases into a single vector, and it is difficult for the model to learn the subtle differences between the cases from long texts. Considering that the content of each part of the case text is relatively fixed, this paper proposes to split the long case text into multiple pieces and encode them separately to obtain the subtle features of different parts. At the same time, learnable affine-transformation is used to improve the similarity scoring module, so that the model learn more subtle differences, which further improves the performance of case matching. The experimental results on the CAIL2019-SCM data set show that compared to another model, the accuracy of the method proposed in this paper have increased by 1.89%.