卷积神经网络在敏感客户模型的应用研究  

Research on Sensitive Clients Model Application Based on Convolutional Neural Networks

作　　者：黎伟健 胡莉琼 朱凯亮 陈钦顺 LI Wei-jian;HU Li-qiong;ZHU Kai-liang;CHEN Qin-shun(China Mobile Internet Co.,ua hou,Guangdong 510640)

机构地区：[1]中移互联网有限公司,广东广州510640

出　　处：《中国质量》2023年第7期107-111,共5页China Quality

摘　　要：研究基于卷积神经网络的文本聚类算法,并在此算法基础上生成敏感客户模型。一是扩大客户声音样本:从10086热线投诉拓展到总部服务标签、智能客服交互记录、满意度短信调研客户反馈、端内用后即评客户声音、互联网社区客户评论等样本。二是运用实时深度学习算法:通过基于卷积神经网络的文本聚类算法实时从客户声音样本中提取客户关注焦点,准确识别敏感客群,并实时添加敏感客户标签。三是拦截手段多样化:拦截能力从单一的营销短信拦截,升级覆盖计费和邮件等生产领域.The research is on text clustering algorithm based on convolutional neural networks,and the sensitive clients model is generated by the algorithm.First,the sample size is enlarged:the sample is expanded from 10086 complaints hotline to clients service labels,chat records of intelligent clients service,clients feedback from SMS and APP investigation of customer satisfaction,clients comment from internet communities.Secondly,real-time deep learning algorithm is applied:extract the clients'focus from clients voice samples by real-time text clustering algorithm based on convolutional neural networks,identify the sensitive clients precisely and add sensitive clients label.Thirdly,interception means are diversified:the interception capability is upgraded from a single marketing SMs interception to cover production fields such as billing and email.

关 键 词：卷积神经网络 文本聚类算法 敏感客户模型 拦截 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] F274[自动化与计算机技术—控制科学与工程] F49[经济管理—企业管理] F626[经济管理—国民经济]