基于分层注意力循环神经网络的司法案件刑期预测

Prison Term Prediction of Judicial Cases Based on Hierarchical Attentive Recurrent Neural Network

为了解决刑期预测任务准确率较差的问题,提出一种基于多通道分层注意力循环神经网络的司法案件刑期预测模型.该模型对传统的循环神经网络模型进行了改进,引入了BERT词嵌入、多通道模式和分层注意力机制,将刑期预测转化为文本分类问题.模型采用分层的双向循环神经网络对案件文本进行建模,并通过分层注意力机制在词语级和句子级两个层面捕获不同词语和句子的重要性,最终生成有效表征案件文本的多通道嵌入向量.实验结果表明:对比现有的基于深度学习的刑期预测模型,本文提出的模型具有更高的预测性能.

In order to solve the problem of poor accuracy of prison term prediction,a prison term prediction model was proposed on the basis of multi-channel hierarchical attentive recurrent neural network.The model improves the traditional recurrent neural network,introduces BERT word embedding,multichannel mode and hierarchical attention mechanism,and transforms the prison term prediction task into text classification problem.The model uses hierarchical bidirectional recurrent neural network to model the legal case text,and captures the importance of different words and sentences at word level and sentence level through hierarchical attention mechanism.Finally,a multi-channel embedding vector that effectively represents the case text is generated.The experimental results show that the proposed model has higher prediction performance compared with the existing prison term prediction model based on deep learning.