摘要
在传统司法领域,刑期判决不可避免地会受到法官主观判断的影响,从而使得在相似案情的情况下判决结果有所不同,甚至极端情况会出现矛盾,即量刑偏差问题。通过大量样本应用神经网络进行刑期预测在一定程度上可以改善量刑偏差的问题,但是由于量刑偏差对数据集质量的影响,从而使得直接使用神经网络进行刑期预测的效果不佳。为减少训练神经网络所需要的大量样本数据以及量刑数据偏差干扰,提出了一种基于先验知识生成虚拟样本与BP神经网络结合进行刑期预测的方法。以预测盗窃罪刑期为对象,在小样本上进行实验,结果证明此方法可以有效改善BP神经网络在小样本刑期预测上的表现,可以使刑期预测相对准确率提升8%,平均绝对误差降低四个月,减少了主观误差对刑期判决的影响,为小样本刑期预测提供一种有效的方法。
In the traditional judicial field,sentence judgment will be inevitably affected by the judge’s subjective judgment,which makes the judgment results different in similar cases and even contradictions in extreme cases,that is,the problem of sentencing deviation.The application of neural network for sentence prediction through a large number of samples can improve the problem of sentencing deviation to a certain extent.However,due to the impact of sentencing deviation on the quality of the data set,the direct use of neural network for the prediction of sentence is not effective.In order to reduce the large amount of sample data required for training the neural network and the bias interference of sentencing data,a method of generating virtual samples based on prior knowledge and combining BP neural network to predict the sentence is proposed.Taking the prediction of the sentence of theft as an object,experiments are performed on a small sample,and the results prove that this method can effectively improve the performance of the BP neural network on the prediction of small sample sentences.It can increase the relative accuracy of the sentence prediction by 8% and reduce the average absolute error by four months,reducing the impact of subjective errors on sentencing sentences and providing an effective method for small sample sentence predictions.
作者
姚直言
赵学龙
戚湧
严悍
YAO Zhi-yan;ZHAO Xue-long;QI Yong;YAN Han(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)
出处
《计算机技术与发展》
2021年第2期138-142,共5页
Computer Technology and Development
基金
国家重点研发计划政府间国际科技创新合作重点专项(2016YFE0108000)
江苏省重点研发计划(产业前瞻与共性关键技术)项目(BE2017163)
南京理工大学本科生科研训练‘千百万’计划立项资助项目(917106840604)。
关键词
先验知识
虚拟样本
小样本
神经网络
刑期预测
prior knowledge
visual sample
small sample
neural network
sentence prediction
