摘要
信通行业客户所提出的交互独立问句缺省较多,且上下文的关联性较强,导致当前的信通客服机器人的回复技术已经无法满足客户需求。为此,提出信通机器人多场景问答的人工智能技术。信通客服机器人回复内部架构的设计包括语音识别、文本识别、语义理解和对话交互等多种功能模块。利用朴素贝叶斯分类器对客户消息进行识别分类。通过联系上下文提取客户消息中的问点,将长短期记忆网络应用在回复消息匹配的计算过程中,以获取较优的回复内容。在客服机器人性能测试中,设置了不同场景以进行测试。测试结果表明,在不同的场景中,信通客服机器人的各个指标均优于传统的客服机器人,并且具有较好的问答回复效果。该结果验证了所设计的客服机器人的场景通用性和有效性。
The interaction-independent questioning proposed by customers in the communication industry has more defaults and strong contextual relevance,which lead the current response techniques of communication customer service robots are no longer able to satisfy the needs of customers.For this reason,artificial intelligence techniques for multi-scenario question and answer of communication robots are proposed.The design of the internal architecture of the communication customer service robot reply includes multiple functional modules such as speech recognition,text recognition,semantic understanding and dialog interaction,etc.The simple Bayesian classifier is used to recognize and classify the customer messages.By commecting the context to extract the question points in the customer messages,the long and short-term memory network is applied in the calculation process of reply message matching to obtain the better reply content.In the customer service robot performance test,different scenarios are set up for testing.The test results show that in different scenarios,the communication customer service robot’s various indicators are all better than the traditional customer service robot,and it has a better question and answer reply effect.The results verify the scenario versatility and effectiveness of the designed customer service robot.
作者
贺辉
张李燕
李博
HE Hui;ZHANG Liyan;LI Bo(Marketing Service Center,State Grid Sichuan Electric Power Company,Chengdu 610000,China)
出处
《自动化仪表》
CAS
2024年第10期12-16,共5页
Process Automation Instrumentation
基金
国网四川科技公司基金资助项目(SGTYHT/21-JS-223)。
关键词
人工智能
信通客服
交互式问句
朴素贝叶斯分类器
长短期记忆网络
多场景问答
消息匹配
Artificial intelligence
Communication customer service
Interactive questioning
Simple Bayesian classifier
Long and short-term memory network
Multi-scenario question and answer
Message matching
