摘要
在客服服务领域,企业要求客服人员使用事先规范的标准服务流程为用户提供相关反馈服务,而由于不同的客服人员业务水平不同,有可能会导致标准服务流程未能被准确执行,影响企业效益和服务质量.所以如何实时对客服人员的实际服务流程与标准服务流程进行一致性检测并对客服人员进行"纠错",成为当前在线客服质检中亟待解决的问题.由于需要对客服的表述进行实时的服务流程挖掘,传统的面向流程模型的一致性验证方法在时效性上无法应用于面向在线客服的服务流程一致性检测场景.口语表达不规范以及词的表述多种多样等问题,也使得一些现有基于关键词匹配的方法不可行.本文将服务流程的一致性检测问题看作基于文本的服务流程序列分类问题,利用有监督的机器学习分类方法予以解决.由于需要对构成服务流程序列的词序进行考虑,本文采用循环神经网络(Recurrent Neural Network,RNN)与卷积神经网络(Convolutional Neural Network,CNN)作为分类模型.考虑到业务初始阶段数据量积累有限以及标注困难,本文针对CNN与RNN结构做出了相应的分析与比较,最后分别得出RNN与CNN在实际数据集下最高的服务流程检测准确度94.55%与92.83%,并且本文分析与比较得出的结论也可为两个模型在实践中的取舍提供一些指导性建议.
In the field of customer service,enterprises require customer service personnel to provide relevant feedback services to users by using standard service processes that are standardized in advance.However,due to the different business levels of customer service personnel,standard service processes are not implemented accurately,thus affecting enterprise benefits and service quality.How to check the consistency between the actual service flow and standard service flow of customer service personnel in real time and correct the errors of customer service personnel has become a problem demanding prompt solution in online customer service quality inspection.Due to the need of real-time service process mining for customer service presentation,the traditional consistency verification method oriented to process model cannot be applied to the online customer service process consistency detection scenario in this paper in terms of timeliness.The non-standard oral expression and the variety of word expression also make some existing methods based on keyword matching not feasible.In this paper,the consistency of the service process detection problem is regarded as the problem of text-based service process sequences classification problem using supervised machine learning classification method to solve.Because of the need to consider the word order that makes up the sequence of service process,this paper decides to adopt Recurrent Neural Network(RNN)and Convolutional Neural Network(CNN)as classification models.In addition,considering the limited amount of data accumulation in the initial phase of the service and the difficulty in labeling,this paper makes corresponding analysis and comparison on the structure of CNN and RNN.Finally,the highest service flow detection accuracy of RNN and CNN under the actual data set is respectively 94.55%and 92.83%,and the conclusions drawn from the analysis and comparison in this paper can also provide some guiding suggestions for the choice between the two models in practice.
作者
莫志强
曹斌
范菁
王俊
MO Zhi-qiang;CAO Bin;FAN Jing;WANG Jun(College of Computer Science&Technology,Zhejiang University of Technology,Hangzhou 310023,China;China Telecom Corporation Limited Zhejiang Branch,Hangzhou 310040,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2022年第2期293-299,共7页
Journal of Chinese Computer Systems
基金
国家重点研发计划项目(2018YFB1402800)资助
浙江省自然科学基金项目(LY19F020030)资助
浙江省大学生科技创新活动计划(新苗人才计划)项目(2019R403092)资助。
