基于过程监督的序列多任务法律判决预测方法

Process Supervision Based Sequence Multi-task Method for Legal Judgement Prediction

法律判决预测是人工智能技术在法律领域的应用,因此对法律判决预测方法的研究对于实现智慧司法具有重要的理论价值和实际意义。传统的法律判决预测方法大都是只进行单一任务的预测或仅基于参数共享的多任务预测,并未考虑各子任务之间的序列依存关系,因此预测性能难以得到进一步的提升。文中提出了一个端到端的基于过程监督的序列多任务法律判决预测模型,在建模各子任务之间的依存关系时,通过引入过程监督来确保依赖信息的准确性,从而提升序列子任务的预测性能。将所提模型应用到CAIL2018数据集上,取得了较好的分类效果,平均分类准确率比现有的state-of-the-art方法的准确率提升了2%。

Legal judgment prediction is an application of artificial intelligence technology in legal field.Hence,the research on the legal judgment prediction method has important theoretical value and practical significance for the realization of intelligent justice.Traditional legal judgment prediction methods only make single task prediction or just use multi-task prediction based on parameter sharing,without considering the sequence dependence among subtasks,so the prediction performance is difficult to be further improved.This paper proposes a process supervision based sequence multi-task framework(PS-SMTL)by encoding sequence dependency of subtasks in legal judgement.It is an end to end legal judgement prediction method without any external features.By introducing process supervision,the proposed model ensures the accuracy of the obtained dependent prior information from advance tasks.The proposed model is applied to CAIL2018 dataset and a good classification result is achieved.The average classification accuracy is 2%higher than that of the existing state-of-the-art method.