ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2022, Vol. 59 ›› Issue (8): 1694-1722.doi: 10.7544/issn1000-1239.20211264

所属专题: 2022数据挖掘前沿进展专题

• 人工智能 • 上一篇    下一篇

基于强化学习的知识图谱综述

马昂1,于艳华1,杨胜利2,石川1,李劼1,蔡修秀1   

  1. 1(北京邮电大学计算机学院(国家示范性软件学院) 北京 100876);2(中国人民解放军国防大学 北京 100091) (ang@bupt.edu.cn)
  • 出版日期: 2022-08-01
  • 基金资助: 
    国家自然科学基金项目(U1936104);国家重点研发计划项目(2020YFB2104503)

Survey of Knowledge Graph Based on Reinforcement Learning

Ma Ang1, Yu Yanhua1, Yang Shengli2, Shi Chuan1, Li Jie1, Cai Xiuxiu1   

  1. 1(School of Computer Science (National Pilot Software Engineering School), Beijing University of Posts and Telecommunications, Beijing 100876);2(National Defence University of People’s Liberation Army, Beijing 100091)
  • Online: 2022-08-01
  • Supported by: 
    This work was supported by the National Natural Science Foundation of China (U1936104), and the National Key Research and Development Program of China (2020YFB2104503).

摘要: 知识图谱是一种用图结构建模事物及事物间联系的数据表示形式,是实现认知智能的重要基础,得到了学术界和工业界的广泛关注.知识图谱的研究内容主要包括知识表示、知识抽取、知识融合、知识推理4部分.目前,知识图谱的研究还存在一些挑战.例如,知识抽取面临标注数据获取困难而远程监督训练样本存在噪声问题,知识推理的可解释性和可信赖性有待进一步提升,知识表示方法依赖人工定义的规则或先验知识,知识融合方法未能充分建模实体之间的相互依赖关系等问题.由环境驱动的强化学习算法适用于贯序决策问题.通过将知识图谱的研究问题建模成路径(序列)问题,应用强化学习方法,可解决知识图谱中的存在的上述相关问题,具有重要应用价值.首先梳理了知识图谱和强化学习的基础知识.其次,对基于强化学习的知识图谱相关研究进行全面综述.再次,介绍基于强化学习的知识图谱方法如何应用于智能推荐、对话系统、游戏攻略、生物医药、金融、安全等实际领域.最后,对知识图谱与强化学习相结合的未来发展方向进行展望.

关键词: 知识图谱, 强化学习, 命名实体识别, 关系抽取, 知识推理, 知识表示, 知识融合

Abstract: Knowledge graph (KG) is a form of data representation that uses graph structure to model the connections between things. It is an important foundation for realizing cognitive intelligence and has received extensive attention from academia and industry. The research of knowledge graph includes four parts: knowledge representation, knowledge extraction, knowledge fusion, knowledge reasoning. Currently, there are still some challenges in the research of knowledge graphs. For example, knowledge extraction methods face difficulty in obtaining labeled data, while distantly supervised training samples have noise problems. The interpretability and reliability of the knowledge reasoning methods need to be further improved. Knowledge representation methods also have problems such as relying on manually defined rules or prior knowledge. Knowledge fusion methods fail to fully model the interdependence between entities. Environment-driven reinforcement learning (RL) algorithms are suitable for sequential decision-making problems. By modeling the research problem of the knowledge graph into a path (sequence) problem, and applying reinforcement learning methods, the above-mentioned problems in the knowledge graph can be solved, which has important application value. The basic knowledge of KG and RL are introduced firstly. Secondly, a research of KG based on RL are comprehensively reviewed. Then, it focuses on how the KG method based on RL can be applied to practical application areas such as intelligent recommendation, conversation system, game, biology, medicine prediction, finance and cybersecurity. Finally, the future directions of KG and RL are discussed in detail.

Key words: knowledge graph, reinforcement learning, named entity recognition, relation extraction, knowledge reasoning, knowledge representation, knowledge fusion

中图分类号: