融入罪名关键词的法律判决预测多任务学习模型

Multi-task learning model for legal judgment predictions with charge keywords

作为新兴的智慧法院技术之一,基于案情描述文本的法律判决预测越来越引起自然语言处理界的关注。罪名预测和法条推荐是法律判决预测的2个重要子任务。这2个子任务密切相关、相互影响,但常常当作独立的任务分别处理。此外,罪名预测和法条推荐还面临易混淆罪名问题。为了解决这些问题,该文提出一种多任务学习模型对这2个任务进行联合建模,同时采用统计方法从案情描述中抽取有助于区分易混淆罪名的指示性罪名关键词,并将它们融入到多任务学习模型中。在CAIL2018法律数据集上的实验结果表明:融入罪名关键词信息的多任务学习模型能够有效解决易混淆罪名问题,并且能够显著地提高罪名预测和法条推荐这2个任务的性能。

The legal field is using more artificial intelligence methods such as legal judgment prediction(LJP)based on case description texts using natural language processing.Charge prediction and law article recommendations are two important LJP sub-tasks that are closely related and interact with each other.However,previous studies have usually analyzed them as two independent tasks that are analyzed separately.Furthermore,charge prediction and law article recommendations both face the problem of confusing charges.To this end,this paper presents a multi-task learning model for joint modeling of charge prediction and law article recommendations.Confusing charges are handled by using a set of charge keywords extracted from case description texts using statistical techniques for integration into the multi-task learning model.This method was evaluated using the CAIL2018 legal dataset.The results show that incorporating the charge keywords into the multi-task learning model effectively resolves the confusing charge problem and significantly improves both the charge prediction and the law article recommendation results.