智能运维场景下的问答系统设计与应用  

Design and application of question answering system in intelligent operation and maintenance scenarios

作　　者：王越 赵艳兴 贺霆 赵逢波 王旭鹏 张治国 WANG Yue;ZHAO Yanxing;HE Ting;ZHAO Fengbo;WANG Xupeng;ZHANG Zhiguo(Beijing Baolande Software Corporation,Beijing 100089,China)

机构地区：[1]北京宝兰德软件股份有限公司,北京100089

出　　处：《智能计算机与应用》2022年第8期54-59,共6页Intelligent Computer and Applications

摘　　要：智能运维(Artificial Intelligence for IT Operations,AIOps)场景中,问答系统的应用可以辅助运维人员完成运维任务,查询运维知识,降低企业运维成本。传统的问答系统功能单一,通常只能完成特定任务或知识查询中的一种。本文设计的多功能问答系统集成了4类问答功能,即:任务型、知识图谱型、问答对型、闲聊型,同时针对任务型场景深度开发Rasa框架,并采用基于BERT的改进型意图识别神经网络结构;通过真实的运维数据验证了该智能问答系统性能。结果表明,本文设计的问答系统不仅可以毫秒级返回问答结果,并且任务型场景中的意图识别准确率在原有基础上有了明显提升。In AIOps scenarios,the application of the question answering system can assist operation and maintenance personnel to complete operation and maintenance tasks,query operation and maintenance knowledge,which can reduce the operation and maintenance cost of enterprises.Usually,traditional question answering systems have a single function and can only complete one of specific tasks or knowledge queries.This paper designs a multifunctional question answering system architecture that integrates four question answering functions(Task,KBQA,FAQ,Chat).At the same time,the Rasa framework is deeply developed for taskbased scenarios,and an improved intent recognition neural network structure based on BERT is adopted;The performance of the intelligent question answering system is verified by real operation and maintenance data.The results show that the question answering system designed in this paper can not only return question and answer results in milliseconds,but also the accuracy of intent recognition in task-based scenarios has been significantly improved on the original basis.

关 键 词：智能运维 问答系统 Rasa框架 意图识别 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]