融合多任务深度学习与主动学习的民航常旅客缺失数据填补  

FILLING MISSING DATA OF CIVIL AVIATION FREQUENT PASSENGERS BY COMBINING MULTI-TASK DEEP LEARNING AND ACTIVE LEARNING

作　　者：李国[1,2] 袁闻 王怀超[1,2] Li Guo;Yuan Wen;Wang Huaichao(School of Computer Science and Technology,China Civil Aviation University,Tianjin 300300,China;Information Technology Research Base Civil Aviation Administration of China,China Civil Aviation University,Tianjin 300300,China)

机构地区：[1]中国民航大学计算机科学与技术学院,天津300300 [2]中国民航大学中国民航信息技术科研基地,天津300300

出　　处：《计算机应用与软件》2022年第8期21-27,共7页Computer Applications and Software

基　　金：民航重要信息系统主动容灾保护关键技术研究项目(U1833114);中央高校基本业务费项目(3122019120);民航科技重大专项(MHRD20160109)。

摘　　要：针对用溯源法主动对数据记录进行真实值填补成本较高的问题,结合多任务学习与主动学习的优势,提出一种深度多任务主动学习网络,将缺失数据填补任务与常旅客价值分类任务同时进行训练。此外,考虑到生产大数据环境,提出一种基于Spark的主动填补运行机制,使得深度多任务网络模型能在大数据环境下高效运行。实验表明,模型在不损失数据填补和分类性能的前提下,模型可有效降低所需训练数据的规模,进而减少训练模型所需时间以及人工标记样本所耗费的人力成本。Aiming at the high cost of using the traceability method to actively fill in the true value of the data records,this paper combines the advantages of multi-task learning and active learning,and proposes a deep multi-task active learning network,which simultaneously fills the missing data and trains the frequent passenger value classification task.Considering the production big data environment,this paper put forward an active filling operation mechanism based on Spark,so as to make the deep multi-task network model run efficiently in the big data environment.Experiments show that the model can effectively reduce the size of the required training data without loss of data filling and classification performance,thereby reducing the time for training the model and the labor cost of manually labeling samples.

关 键 词：民航常旅客运营数据 缺失值 常旅客价值分类 深度多任务主动学习 降噪自编码器 SPARK 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]