摘要
方案提出一种基于SpanBERT(Bidirectional Encoder Representations from Transformers by representing and predicting Spans)模型的服务热线文本情感分析方法,以SpanBERT实现句向量优化的文本情感细粒度分析方案,针对移动客服与用户对话数据,实现场景化客服文本分析,通过挖掘负面投诉对话文本价值,并基于识别的客户情绪、语义信息等进行质检,可提前获知客户的潜在不满意倾向,持续提高客户的服务体验,具有很好的推广前景。已应用在天津移动满意度预测、服务运营分析和语音质检工作中,以投诉语音质检机器人替代人工操作,实现降本增效。
A service voice negative emotion recognition method based on the SpanBERT(Bidirectional Encoder Representations from Transformers by representing and predicting Spans)model is proposed,which uses SpanBERT to achieve sentence vector optimization for fine-grained text sentiment analysis.Based on conversation data between customer and cervicer,scenario based customer service text analysis is implemented.By mining the value of negative complaint dialogue text and conducting quality inspection based on identified customer emotions,semantic information,etc.,potential dissatisfaction tendencies of customers can be identified in advance,Continuously improving customer service experience has good promotion prospects.It has been applied in China Mobile’s satisfaction prediction,service operation analysis,and voice quality inspection work to complain about voice quality inspection robots,replacing manual operations,and achieving cost reduction and efficiency improvement.
作者
赵东明
张继军
王博
张亚洲
刘静
石理
ZHAO Dong-ming;ZHANG Ji-jun;WANG Bo;ZHANG Ya-zhou;LIU Jing;SHI Li(Artificial Intelligence Laboratory of China Mobile Communications Group Tianjin Co.,Ltd,Tianjin 300020,China;Intelligence and Computing Department,Tianjin University,Tianjin 300192,China)
出处
《新一代信息技术》
2023年第15期1-5,共5页
New Generation of Information Technology
基金
国家自然科学基金(No.62376188)
关键词
情感计算
SpanBERT模型
自然语言理解
负面情绪识别
语义理解
多模异态
emotional computing
SpanBERT model
natural language understanding
multimodal emotion recognition
semantic understanding
multimodal heteromorphism
