基于知识图谱的多轮对话技术研究综述  

Recovery of Multi-turn Dialogue Based on Knowledge Graph

作　　者：杨阳 盛胜利 奚雪峰 YANG Yang;SHENG Sheng-li;XI Xue-feng(School of Electronic and Information Engineering,Suzhou University of Science and Technology,Suzhou 215009,China;Suzhou Key Laboratory of Virtual Reality Intelligent Interaction and Application Technology,Suzhou University of Science and Technology,Suzhou 215009,China;Data Analytics Lab,Soochow University(Texas Tech University),Lubbock 79409,USA)

机构地区：[1]苏州科技大学电子与信息工程学院,江苏苏州215009 [2]苏州市虚拟现实智能交互及应用重点实验室(苏州科技大学),江苏苏州215009 [3]数据分析实验室(德州理工大学),德克萨斯州拉伯克市79409

出　　处：《计算机技术与发展》2023年第4期27-33,共7页Computer Technology and Development

基　　金：国家自然科学基金项目(61876217,62176175);江苏省“六大人才高峰”高层次人才项目资助(XYDXX-086)。

摘　　要：随着自然语言技术的不断进步与发展,人机交互取得了跨越式的进步。然而,目前人机交互系统往往都是用户与机器双方在特定的应用场景下设计完成的,在开放域下进行难度较大的多轮对话效果差强人意。而知识图谱作为实现对话系统的重要工具之一,其被证明在多轮对话任务中是有效的。该文从基于知识图谱的多轮对话技术总结了多轮对话中使用的相关技术,其中基于知识图谱的多轮对话模型包括TransE、TransH、TransR和TransD等,以及涉及到基于知识图谱的多轮对话相关数据集及评价标准。最后提出了基于知识图谱的多轮对话技术当前面临的挑战并进行了总结。With the continuous progress and development of natural language technology,human-computer interaction has made a great progress.However,at present,human-computer interaction systems are often designed by both users and machines in specific application scenarios,and the effect of difficult multi-turn dialogue in open domain is not satisfactory.As one of the important tools to implement dialogue system,knowledge graph has been proved to be effective in multi-turn dialogue tasks.We summarize the related technologies used in multi-turn dialogue from the multi-turn dialogue technology based on knowledge graph.The multi-turn dialogue model based on knowledge graph includes TransE,TransH,TransR and TransD,as well as the related data sets and evaluation standards of multi-turn dialogue based on knowledge graph.Finally,we put forward the current challenges of multi-turn dialogue technology based on knowledge graph and make a summary.

关 键 词：知识图谱 多轮对话 人机交互 自然语言处理 对话系统 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]