Please wait a minute...
浙江大学学报(工学版)  2021, Vol. 55 Issue (1): 109-115    DOI: 10.3785/j.issn.1008-973X.2021.01.013
计算机技术、自动控制技术     
基于深度神经网络的多因素感知终端换机预测模型
陈纬奇1(),王敬昌2,陈岭1,*(),杨勇勤3,吴勇2
1. 浙江大学 计算机科学与技术学院,浙江 杭州 310027
2. 浙江鸿程计算机系统有限公司,浙江 杭州 310009
3. 中国电信浙江分公司,浙江 杭州 310040
Prediction model of multi-factor aware mobile terminal replacement based on deep neural network
Wei-qi CHEN1(),Jing-chang WANG2,Ling CHEN1,*(),Yong-qin YANG3,Yong WU2
1. College of Computer Science and Technology, Zhejiang University, Hangzhou 310027, China
2. Zhejiang Hongcheng Computer Systems Limited Company, Hangzhou 310009, China
3. Zhejiang Branch of China Telecom Limited Company, Hangzhou 310040, China
 全文: PDF(889 KB)   HTML
摘要:

针对基于特征工程的传统终端换机预测模型依赖于领域知识且无法充分利用用户通话、流量使用等序列数据的问题,提出基于深度神经网络的多因素融合终端换机预测模型. 该模型使用长短时记忆网络(LSTM)提取用户通话、流量使用行为序列特征,使用全连接网络融合用户自然属性、行为序列特征和历史换机信息,预测用户是否换机. 实验表明,基于深度神经网络的多因素融合终端换机预测模型能够考虑影响用户换机的多种因素,充分挖掘用户通话、流量使用行为序列特征;当召回率为0.135时,相比于传统模型精确率提高了34.3%.

关键词: 终端换机预测多因素感知深度神经网络长短时记忆网络全连接网络    
Abstract:

A multi-factor aware mobile terminal replacement prediction model based on deep neural networks was proposed to address the problem that traditional mobile terminal replacement prediction models based on feature engineering rely on the domain knowledge and cannot sufficiently use user’s call details and data traffic details. Long short-term memory (LSTM) networks were utilized to extract the sequence characteristics of user’s call and data traffic behaviors. Then a fully connected neural network was utilized to fuse user’s natural attributes, sequence characteristics, and historical terminal replacement information for prediction. The experimental results show that the proposed model can consider multiple factors affecting terminal replacement and sufficiently exploit the sequence characteristics of user’s call details and data traffic details. The precision was increased by 34.3% compared with traditional methods when recall was set to 0.135.

Key words: mobile terminal replacement prediction    multi-factor aware    deep neural network    long short-term memory network    fully connected neural network
收稿日期: 2019-12-18 出版日期: 2021-01-05
CLC:  TP 491  
基金资助: 国家重点研发计划资助项目(2018YFB0505000);中央高校基本科研业务费专项资金资助项目(2020QNA5017)
通讯作者: 陈岭     E-mail: vc12301@gmail.com;lingchen@cs.zju.edu.cn
作者简介: 陈纬奇(1995—),男,硕士生,从事数据挖掘的研究. orcid.org/0000-0001-9478-4683. E-mail: vc12301@gmail.com
服务  
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章  
陈纬奇
王敬昌
陈岭
杨勇勤
吴勇

引用本文:

陈纬奇,王敬昌,陈岭,杨勇勤,吴勇. 基于深度神经网络的多因素感知终端换机预测模型[J]. 浙江大学学报(工学版), 2021, 55(1): 109-115.

Wei-qi CHEN,Jing-chang WANG,Ling CHEN,Yong-qin YANG,Yong WU. Prediction model of multi-factor aware mobile terminal replacement based on deep neural network. Journal of ZheJiang University (Engineering Science), 2021, 55(1): 109-115.

链接本文:

http://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2021.01.013        http://www.zjujournals.com/eng/CN/Y2021/V55/I1/109

图 1  换机行为与用户特征相关性
图 2  终端换机预测模型框架
类别 属性 说明
用户自然属性 性别 男/女
年龄 0~100 岁
入网时长 0~240 个月
是否为亲情网用户 是/否
同客户下C网个数 0~20 个
App类型偏好 社交、视频、阅读等
终端属性 品牌 华为、苹果、小米等
价格 0~15000 元
屏幕尺寸 0~6.95 英寸
网别 3G/4G
像素 0~4000万像素
表 1  用户自然属性和终端属性说明
图 3  终端品牌使用分布
图 4  LSTM网络结构
网络层 超参数
LSTM层 units=d
全连接层 第1层:units=64;Activation=LeakyReLU( $ \alpha $=0.1)
第2层:units=32;Activation=LeakyReLU( $ \alpha $=0.1)
第3层:units=1;Activation=Sigmoid
表 2  模型超参数设置
图 5  LSTM输出维度d的影响
方法 P R F1
MRPM-U 0.165 0.135 0.148
MRPM-S 0.157 0.135 0.145
MRPM-H 0.103 0.135 0.117
MRPM 0.192 0.135 0.159
表 3  忽略不同因素对模型性能的影响
数据集 统计特征
用户通话数据 近30 日日平均通话次数
近180 日日平均通话次数
近30 日日平均通话时长
近180 日日平均通话时长
近30 日当日较上日通话次数平均增长率
近180 日当日较上日通话次数平均增长率
近30 日当日较上日通话时长平均增长率
近180 日当日较上日通话时长平均增长率
用户流量使用数据 近30 日日平均上行流量
近180 日日平均上行流量
近30 日日平均下行流量
近180 日日平均下行流量
近30 日当日较上日上行流量平均增长率
近180 日当日较上日上行流量平均增长率
近30 日当日较上日下行流量平均增长率
近180 日当日较上日下行流量平均增长率
表 4  用户行为数据统计特征
方法 P R F1
LR 0.105 0.135 0.118
SVM 0.121 0.135 0.128
XGBoost 0.143 0.135 0.139
MRPM 0.192 0.135 0.159
表 5  召回率为0.135时不同模型的精确率和F1
图 6  不同模型的精确率-召回率曲线
1 ZHANG K, CAI Z, LV T. The research on the pricing system of mobile value-added service for 3G [C]//International Conference on Management Science and Engineering. Harbin: IEEE, 2007: 234-238.
2 CHEN J, GUO J. A choreography approach for value-added services creation [C]//International Conference on Wireless Communications, Networking and Mobile Computing. Beijing: IEEE, 2009: 1-4.
3 BUGHIN J Reaping the benefits of big data in telecom[J]. Journal of Big Data, 2016, 3 (1): 14
doi: 10.1186/s40537-016-0048-1
4 ZHAO Z, GUO J, DING E, et al. Terminal replacement prediction based on deep belief networks [C]//International Conference on Network and Information Systems for Computers. Wuhan: IEEE, 2015: 255-258.
5 HADDEN J, TIWARI A, ROY R, et al Churn prediction: does technology matter[J]. International Journal of Intelligent Technology, 2006, 1 (2): 104- 110
6 SHAO G, CHEN W, CHENG X, et al. Telecom big data based user analysis and application in telecom industry [C]//International Conference on 5G for Future Wireless Networks. Beijing: Springer, 2017: 99-109.
7 ZHANG Z, LIN H, LIU K, et al A hybrid fuzzy-based personalized recommender system for telecom products/services[J]. Information Sciences, 2013, 235: 117- 129
doi: 10.1016/j.ins.2013.01.025
8 YANG D, WU Z, WANG X, et al. Predicting replacement of smartphones with mobile App usage [C]//International Conference on Web Information Systems Engineering. Shanghai: Springer, 2016: 343-351.
9 HUANG B, KECHADI M, BUCKLEY B Customer churn prediction in telecommunications[J]. Expert Systems with Applications, 2012, 39 (1): 1414- 1425
doi: 10.1016/j.eswa.2011.08.024
10 VAEFIADIS T, DIAMANTARAS K, SARIGIANNIDIS G, et al A comparison of machine learning techniques for customer churn prediction[J]. Simulation Modelling Practice and Theory, 2015, 55: 1- 9
doi: 10.1016/j.simpat.2015.03.003
11 XIA G, JIN D Model of customer churn prediction on support vector machine[J]. Systems Engineering: Theory and Practice, 2008, 28 (1): 71- 77
doi: 10.1016/S1874-8651(09)60003-X
12 IDRIS A, ASIFULLAH K, YEON S. Genetic programming and adaboosting based churn prediction for telecom [C]//International Conference on Systems, Man, and Cybernetics. Seoul: IEEE, 2012: 1328-1332.
13 JAMIL S, ASIFULLAH K. Churn comprehension analysis for telecommunication industry using ALBA [C]// International Conference on Emerging Technologies. Islamabad: IEEE, 2016: 1-5.
14 LAURILA J, GATICA-PEREZ D, AAD I, et al From big smartphone data to worldwide research: the mobile data challenge[J]. Pervasive and Mobile Computing, 2013, 9 (6): 752- 771
doi: 10.1016/j.pmcj.2013.07.014
15 STONE C. Optimal global rates of convergence for nonparametric regression [J]. The Annals of Statistics. 1982, 10(4): 1040-1053.
16 LIAN J, ZHOU X, ZHANG F, et al. xDeepFM: combining explicit and implicit feature interactions for recommender systems [C]//ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. London: ACM, 2018: 1754-1763.
17 SUTSKEVER I, VINYALS O, LE Q. Sequence to sequence learning with neural networks [C]//Advances in Neural Information Processing Systems. Montreal: MIT Press, 2014: 3104-3112.
18 HOCHREITER S, SCHMIDHUBER J Long short-term memory[J]. Neural Computation, 1997, 9 (8): 1735- 1780
doi: 10.1162/neco.1997.9.8.1735
19 熊冰妍, 王国胤, 邓维斌 分级式代价敏感决策树及其在手机换机预测中的应用[J]. 山东大学学报: 工学版, 2015, 45 (5): 36- 42
XIONG Bing-yan, WANG Guo-yin, DENG Wei-bin Hierarchical cost sensitive decision tree and its application in the prediction of the mobile phone replacement[J]. Journal of Shandong University: Engineering Science, 2015, 45 (5): 36- 42
[1] 许佳辉,王敬昌,陈岭,吴勇. 基于图神经网络的地表水水质预测模型[J]. 浙江大学学报(工学版), 2021, 55(4): 601-607.
[2] 毛奕喆,龚国芳,周星海,王飞. 基于马尔可夫过程和深度神经网络的TBM围岩识别[J]. 浙江大学学报(工学版), 2021, 55(3): 448-454.
[3] 陈世达,刘强,韩亮. 降低分布式训练通信的梯度稀疏压缩方法[J]. 浙江大学学报(工学版), 2021, 55(2): 386-394.
[4] 江金生,任浩然,李瀚野. 基于卷积自编码器的地震数据处理[J]. 浙江大学学报(工学版), 2020, 54(5): 978-984.
[5] 王昶,张永生,王旭,于英. 基于深度学习的遥感影像变化检测方法[J]. 浙江大学学报(工学版), 2020, 54(11): 2138-2148.
[6] 王敬昌,陈岭,余珊珊,蒋晨书,吴勇. 基于门控循环单元的多因素感知短期游客人数预测模型[J]. 浙江大学学报(工学版), 2019, 53(12): 2357-2364.