融合BERT语义特征的社区矫正方案推荐技术  

Community correction scheme recommendation technique combining BERT semantic features

作　　者：柳阳 郭红钰[1] LIU Yang;GUO Hong-yu(Dept.8 of System,North China Institute of Computing Technology,Beijing 100083,China)

机构地区：[1]华北计算技术研究所系统八部,北京100083

出　　处：《电子设计工程》2020年第6期1-5,共5页Electronic Design Engineering

基　　金：国家科技部重点项目(司法专项)(2018YFC0831103)。

摘　　要：针对社区矫正人员繁多且情况愈加复杂,目前人力难以实现社区矫正方案实时精准推送的现实问题。本论文就某市司法局采集到的社区矫正人员状态信息数据,结合实际业务需求,通过采用BERT预训练特征向量分别在GBDT和XGBoost模型上训练实现方案推荐,并对比传统onehot编码方法,比较迁移学习改进后模型的优越性。模型最终测试性能达到93%。此外,针对实际的司法矫正人员特征情况提出最优的特征选择方法。In view of the large number of community corrections personnel and the increasingly compli-cated situation,it is difficult for human resources to realize the real-time and accurate push of community correction programs in real time.This paper analyzes the status information of community correctors col-lected by a city’s judicial bureau,and combines the actual service requirements with the BERT pre-train-ing feature vector to implement the scheme recommendation on the GBDT and XGBoost models,and com-pares the traditional one-hot coding method.Compare the superiority of the migration learning improved model.The final test performance of the model reached 93%.In addition,an optimal feature selection method is proposed for the actual situation of judicial correction personnel.

关 键 词：社区矫正方案推荐 迁移学习 BERT GBDT XGBoost 

分 类 号：TN081[电子电信—物理电子学]