图书情报工作 ›› 2022, Vol. 66 ›› Issue (5): 125-132.DOI: 10.13266/j.issn.0252-3116.2022.05.013

• 知识组织 • 上一篇    下一篇

融合多层次数据的问答知识图谱本体模型构建

周毅1, 刘峥1,2, 粟小青3, 金体成3   

  1. 1. 中国科学院文献情报中心 北京 100190;
    2. 中国科学院大学经济与管理学院图书情报与档案管理系 北京 100190;
    3. 华为终端有限公司 深圳 518129
  • 收稿日期:2021-07-06 修回日期:2021-11-05 出版日期:2022-03-05 发布日期:2022-03-21
  • 通讯作者: 刘峥,研究馆员,博士,通信作者,E-mail:liuz@mail.las.ac.cn
  • 作者简介:周毅,馆员,硕士;粟小青,知识管理专家,硕士;金体成,知识管理工程师。
  • 基金资助:
    本文系国家重点研发计划项目“先进制造业分布式科技服务技术集成研发与示范”(项目编号:2019YFB1405100)研究成果之一。

Ontology Model Construction of Question-Answering Knowledge Graph Integrating Multi-Level Data

Zhou Yi1, Liu Zheng1,2, Su Xiaoqing3, Jin Ticheng3   

  1. 1. National Science Library, Chinese Academy of Sciences, Beijing 100190;
    2. Department of Library, Information and Archives Management, School of Economics and Management, University of Chinese Academy of Sciences, Beijing 100190;
    3. Huawei Device Co. Ltd., Shenzhen 518129
  • Received:2021-07-06 Revised:2021-11-05 Online:2022-03-05 Published:2022-03-21

摘要: [目的/意义] 针对基于问答对的智能问答准确率和解决率低、用户满意度差等问题,研究构建知识图谱本体模型,构建基于知识图谱的智能问答,解决基于问答对的智能问题所面临的问题。[方法/过程] 首先,分析当前智能问答面临的问题及原因,提出构建知识图谱支撑智能问答的方案。其次,在已有本体模型构建方法的基础上,提出一种融合多层次数据的多轮循环方法,该方法分别以业务数据、用户数据和业务系统动态数据等多层次数据为数据来源,核心步骤为搭建基本框架、完善知识结构、对齐知识结构三轮循环。最后,以退换货领域为例阐述本体模型构建的具体步骤,从无到有,增量叠加,构建知识图谱本体模型。[结果/结论] 将以退换货本体模型为模式层的知识图谱部署在智能问答系统中进行试验,试验结果显示退换货知识图谱上线后智能问答的准确率提升50%,解决率提升300%。其中准确率是指回答正确的问题数量与回答的全部问题数量的比例,解决率是指答案精准解决了用户问题的数量与回答的全部问题数量的比例。本文提出的本体模型构建方法从零散的领域知识中梳理出完整的、细粒度的领域知识结构,支持智能问答为用户提供精准的答案,能够有效解决基于问答对的智能问答困境。

关键词: 知识图谱, 本体模型, 精准问答, 多层次数据

Abstract: [Purpose/significance] Aiming at problems of intelligent Q&A based on Q&A pairs such as low accuracy and resolution rate and poor user satisfaction, this paper constructs a knowledge graph (KG) ontology model that supports the realization of dynamic and accurate intelligent Q&A based on the knowledge graph. [Method/process] First, the paper analyzed the current problems and causes of intelligent question answering, and proposed a plan to build a knowledge graph to support intelligent question answering. Second, On the basis of existing ontology model construction methods, the paper proposed a multi-round loop method integrating multi-level data, which used the business data provided by the enterprises, user data and business system dynamic data as the data sources. And the core steps were to build a basic framework, improve the knowledge structure, and align three cycles of the knowledge structure. Finally, this paper took the domain of return and exchange as a case to describe the concrete steps of ontology model construction, from zero, added incrementally, and constructed ontology model of knowledge graph. [Result/conclusion] This paper applies the knowledge graph with the return ontology model as the schema layer in an intelligent Q&A system for testing. The evaluation results show that the accuracy rate increased by 50% and the precision rate increased by 300% after the return and exchange knowledge graph is online. So, the proposed ontology model construction method sorts out the complete and fine-grained domain knowledge structure from scattered domain knowledge, can provide accurate answers to users in intelligent Q&A, and can effectively solve the intelligent Q&A dilemma based on Q&A pairs.

Key words: knowledge graph, ontology model, accurate Q&A, multi-level data

中图分类号: