基于XLCLS模型的法条多标签分类方法研究

为了强化案件语义信息以及法条间的联系，提升多法条自动推送效果，将法条司法解释加入到模型训练中，提出了一种基于XLCLS（XLNET-CNN Model Integrating Legal Interpretation and Similarity，融合法律解释和相似性的XLNET-CNN模型）模型的法条多标签分类方法。该方法首先利用数据增强对案件进行扩充预处理；引入XLNET模型提升案件的向量化表征能力，再加入法条司法解释加强案件语义信息；然后通过CNN模型获取特征信息，进而利用法条间相关性来解决罕见法条容易被忽略的问题，并结合sigmoid函数计算各法条概率；最后通过阈值的选取来实现法条多标签分类。实验结果表明，在公益诉讼案件数据上，该方法的F1,ma与F1,mi上分别达到了88.37%与89.92%，验证了所提方法的可行性与有效性。

Research on multi-label classification method of law for public interest litigation cases

In order to strengthen the connection between semantic information of cases and laws, and improve the effect of automatic push of multiple laws, A Model based on XLCLS (XLNET-CNN Model Integrating Legal Interpretation and Similarity) was proposed by adding judicial interpretation of laws into model training.Firstly, data enhancement is used to preprocess the case;XLNET model was introduced to improve the representation ability of cases,and judicial interpretation was added to strengthen the semantic information of cases.Then, feature information is obtained by CNN model, and the correlation between laws is used to solve the problem that rare laws are easy to be ignored, and the probability of each law is calculated by sigmoid function. Finally, the multi-label classification of law is realized by selecting threshold value. Experimental results show that the F1,ma and F1,mi of the proposed method reach 88.37% and 89.92% respectively in the data of public interest litigation cases, which verifies the feasibility and effectiveness of the proposed method.