一个新的上下文感知类案匹配与推荐方法

A Novel Context-aware Similar Case Matching and Recommendation Method

根据现有的类案推荐方法得到的相似案件,其精确度通常很难满足法官的需求,辅助判案效果有限。为此,提出了一个基于上下文感知的类案匹配和推荐模型(CASCMR).模型为实现端到端高效率的文本匹配与推荐,使用多语义文档表达框架,并通过文本向量预先计算与存储,减少匹配时间,进而提高效率。具体而言,为了更好地对法律长文本进行建模,CASCMR使用BERT对数据进行编码,利用其注意力机制捕获文本长距离依赖信息。同时,考虑法律文本从局部到全局的信息,利用双向LSTM和CNN分别获取文本的上下文信息和局部语义特征,提高法律文本的表征能力,从而提升模型的预测性能。将所提出的模型应用到中国法研杯2019相似案件匹配任务,实验结果显示,与目前最好的方法相比,匹配和推荐精度的提升效果较为明显。

As far as we know,it is usually difficult for the accuracy of similar cases obtained by the existing case recommendation methods to meet the needs of judges,and thus the effect of auxiliary judgment is limited.Therefore,in this paper a novel context-aware similar case matching and recommendation model(CASCMR).In order to achieve end-to-end efficient text matching and recommendation was proposed.The model uses a multi-semantic document expression framework to realize the pre-calculation and storage of text vectors,so as to reduce the matching time and improve the efficiency.Specifically,in order to model the long legal text,CASCMR uses BERT for encoding since its attention mechanism can capture the long-term dependency well.At the same time,the global and local information of the legal text as captured by Bi-LSTM and CNN,respectively,is considered to be helpful to improve the representation of the text,as well as the prediction performance of the model.The proposed model was then applied to the similar case matching task of CAIL2019-SCM,and its accuracy is higher than that of the state-of-the-art method.