三元组深度哈希学习的司法案例相似匹配方法

Triplet deep Hashing learning for judicial case similarity matching method

在数量庞大的司法案例文书中进行相似案例匹配可以有效地提升司法部门的工作效率。但司法案件文本不仅长,而且文本自身还具有一定程度的结构复杂性,因此司法案例文本匹配与传统自然语言处理任务相比,具有较高的难度。为解决上述问题,本文基于三元组深度哈希学习模型提出了一种司法案例相似匹配方法,首先使用预训练的BERT中文模型分组提取文书的特征;再利用文书三元组相似性关系,训练深度神经网络模型,用于生成文书的哈希码表示;最后,基于文书哈希码的海明距离来判断是否为相似案例。实验结果表明,本文采用哈希学习方法极大地降低了文书特征表示的存储开销,提高了相似案例匹配的速度。

Matching similar cases in a large number of judicial case documents can effectively improve the efficiency of the judicial department.However,the text of judicial cases is not only lengthy,but also exhibits a certain degree of structural complexity.Therefore,the text matching of judicial cases is more difficult compared with the traditional natural language processing tasks.To solve the above problems and challenges,this paper proposes a judicial case similar matching method based on the triplet deep Hashing learning model.First,a pre-trained BERT model is used to extract the features of the documents in groups.The triplet similarity relationship of the documents is then employed to train the deep neural network model to generate the Hashing code representation of the documents.Finally,the Hamming distance based on the Hashing code of the documents is used to determine whether they are similar cases.Experimental results show that the Hashing learning method greatly reduces the storage cost of the documents’feature representations and improves the speed of similar case matching.