摘要
在法律判决预测领域中,案件描述通常具有相似的结构,而现有的预测方法容易忽略不同案件间的要素差异,难以有效利用案件要素特征,导致模型预测准确率不高.此外,罪名预测任务还面临易混淆罪名问题.针对上述问题,提出了一种融合案件要素和案件属性的罪名预测多任务学习模型(Case Elements And Attributes Multi-Task Learning Model,简称CEAT-MLM),通过挖掘案件要素及案件属性与罪名之间的关联关系,将案件属性预测和罪名预测进行联合建模,达到提升罪名预测准确率的目标.实验结果表明,本文提出的模型相较于通用的文本分类模型具有显著的性能提升,并与法律判决领域的典型模型相比,Macro-F1得分提升了1.76%.
In the field of legal judgment prediction,case descriptions usually have similar structures,and existing prediction methods tend to ig-nore the differences in elements between different cases,making it difficult to effectively utilize the characteristics of case elements,resulting in low accuracy of model predictions.In addition,the task of predicting charges also faces the problem of confusing charges.A Case Elements And Attributes Multi Task Learning Model(CEAT-MLM)that integrates case elements and attributes is proposed to address the above issues.By mining the correlation between case elements and attributes and charges,the case attribute prediction and charge prediction are jointly modeled to achieve the goal of improving the accuracy of charge prediction.The experimental results show that the proposed model has significant performance improvement compared to general text classification models,and compared with typical models in the field of legal judgment,the Macro-F1 score has increased by 1.76%.
作者
李为祖
武友新
于程远
LI Weizu;WU Youxin;YU Chengyuan(School of Mathematics and Computer Sciences,Nanchang University,Nanchang 330031,China;School of Computer Science&Engineering,Jiangxi Agricultural University,Nanchang 330045,China)
出处
《小型微型计算机系统》
北大核心
2026年第1期106-112,共7页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(62163018)资助
江西省03专项及5G基金项目(20193ABC03A010)资助。
关键词
罪名预测
案件要素
易混淆罪名
多任务学习
predicting charges
case elements
confusing charges
multi-task learning
