基于AlBert-Tiny-DPCNN的案件事实倾向性类别预测

Chinese judicial document processing by utilizing case fact label classification based on AlBert-Tiny-DPCNN

近年来,随着智慧司法的推进,中国裁判文书作为重要的研究对象,衍生了诸多任务,但针对裁判文书的研究大多基于刑事案件,缺乏对民事案件领域下的研究。本文结合预训练词向量、文本分类模型等技术,对民间借贷这一细分领域下的案件事实标签预测进行研究,为现有的案件事实提供同类别的有参考价值的裁判文书,减少相关工作者在大量数据中寻找所耗费的时间。本文提出了基于Albert-Tiny-DPCNN的分类模型,该模型采用注意力机制与标签平滑归一化技术来提高模型的精度,并在实验数据集上验证了模型的有效性。

In recent years,with the advancement of intelligent justice,judicial documents in China,as an important research object,have derived many tasks.However,most of the research on judicial documents is based on criminal cases,and there is a lack of research on civil cases.Combining pre-trained word vectors,text classification models,and other technologies,this paper studies the prediction of case facts in the subdivision field of private lending so as to provide the existing case facts with similar judgment documents with reference value and reduce the time spent by relevant workers looking for them in a large amount of data.In this paper,a classification model based on Albert-Tiny-DPCNN is proposed,attention mechanisms and label smoothing regularization techniques are used to improve the accuracy of the model,and the validity of the model is verified on experimental data sets.