摘要
自2018年以来,铁路客服中心开始引入人工智能技术,并在语音机器人场景进行系统试用,然而试用过程中发现存在知识检索精度低以及知识构建和维护效率低等问题。为了解决上述问题,文章主要研究了基于铁路客运客服知识库的意图识别模型训练及应用技术,从以下三个方面展开探讨:首先,介绍了多预训练模型融合蒸馏的意图识别模型训练技术,包括模型预训练、模型融合和模型蒸馏等关键技术;其次,研究了基于意图识别的信息检索技术,该技术在传统基于关键字的信息检索技术的基础上,结合了检索意图识别,从而提高了知识检索的精度;最后,介绍了基于规则与模型结合的非结构化知识自动采编技术,通过设计规则抽取和意图识别模型抽取两种方法,实现了从海量非结构化文本中自动抽取和采编结构化知识的功能,从而解决了知识构建及维护效率低的问题。
Since 2018,the railway customer service center has begun to introduce artificial intelligence technology and conducted system trial in the voice robot scene.During the trial process,it was found that there were issues such as low accuracy in knowledge retrieval and low efficiency in knowledge construction and maintenance.In order to solve the above problems,this paper mainly studies the intention recognition model training and application technology based on the railway passenger service knowledge base and explores three aspects:intent recognition model training,information retrieval based on intent recognition,and automatic collection and compilation of unstructured knowledge.Firstly,introduce the intention recognition model training technology of multi-pre-training model fusion distillation,including model pre-training,model fusion,model distillation,and other technologies.Secondly,explore the information retrieval technology based on intent recognition,and improve the accuracy of knowledge retrieval by combining retrieval intent recognition on the basis of traditional keyword-based information retrieval technology.Finally,introduce the unstructured knowledge automatic collection and editing technology based on the combination of rules and models.Through the two methods of design rule extraction and intent recognition model extraction,the function of automatically extracting and editing structured knowledge from massive unstructured texts is realized,and the problem of low efficiency in knowledge construction and maintenance is solved.
作者
杨立鹏
陈华龙
张睿
王勃
孟庆晔
李强
YANG Lipeng;CHEN Huaong;ZHANG Rui;WANG Bo;MENG Qingye;LI Qiang(Beijing Jingwei Information Technology Co.,Ltd.,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;iFLYTEK AI Research(Hebei),LangFang,Hebei 065000,China;Iflytek(Beijing)Co.,Ltd.,Beijing 100000,China;Beijing Railway Customer Service Center,China Railway Beijing Group Co.,Ltd.,Beijing 100860,China)
出处
《计算机应用文摘》
2023年第21期49-54,共6页
Chinese Journal of Computer Application
基金
中国国家铁路集团有限公司科技研究开发计划课题(N2022S002)。
关键词
铁路客服
自然语言处理
意图识别
信息检索
非结构化知识采编
railway passenger service
natural language processing
intent recognition
information retrieval
automatic collection and compilation of unstructured knowledge
