家庭服务机器人领域知识图谱构建与应用-《计算机技术与发展》

文章信息/Info

Title:: Domain Knowledge Graph Construction and Application of Home Service Robot

作者:: 吴培良¹; 2 ; 王天成¹; 2 ; 金鑫龙¹; 2 ; 闫鹏宇³ ; 张云川¹; 2 ; 陈雯柏⁴ ; 毛秉毅¹; 2 ; 高国伟⁴; 1. 燕山大学信息科学与工程学院,河北秦皇岛 066004;
2. 河北省计算机虚拟技术与系统集成重点实验室,河北秦皇岛 066004;
3. 上海工业自动化仪表研究院有限公司,上海 200233;
4. 北京信息科技大学自动化学院,北京 100192

Author(s):: WU Pei-liang1; 2 ; WANG Tian-cheng1; 2 ; JIN Xin-long1; 2 ; YAN Peng-yu3 ; ZHANG Yun-chuan1; 2 ; CHEN Wen-bai4 ; MAO Bing-yi1; 2 ; GAO Guo-wei4; 1. School of Information Science and Engineering,Yanshan University,Qinhuangdao 066004,China;
2. Key Laboratory for Computer Virtual Technology and System Integration of Hebei Province,Qinhuangdao 066004,China;
3. Shanghai Industrial Automation Instrument Research Institute,Shanghai 200233,China;
4. School of Automation,Beijing Information Science and Technology University,Beijing 100192,China

Keywords:: home service robot; service policy; semantic information; knowledge graph; deep learning

摘要:: 对于家庭服务机器人,能否准确快速地获取到家庭场景中实体的语义信息是决定其智能化水平的关键。为了增强其语义信息获取能力与知识推理能力,针对家
庭场景提出了一种面向服务机器人的领域知识图谱自动化构建流程。首先,利用词频-逆向文件频率算法(Term Frequency – Inverse Document Frequency,TF-IDF) 从文本信息中提取服务策略关键字,构建服务策略图谱;其次,通过预训练的场景分割模型( Scene Segmentation Model,SSM) 识别出场景内的实体;之后,根据当前场景的实体信息,运用场景分类模型(Scene Classification Model,SCM) 来预测房间类别,生成结构化数据;再次,根据结构化数据构建家庭实体图谱;最后,
将服务策略图谱与家庭实体图谱合并为家庭服务领域知识图谱,并将其存储到 neo4j 图数据库中。实验结果表明,所提出的方法可以根据非结构化数据自动生成领域知识图谱,通过查询知识图谱,检索家庭服务所需的语义信息,可以帮助机器人生成适用于当前工作环境的服务策略,使其更加智能地完成服务任务,证明了方法的可行性。

Abstract:: For the home service robot,whether it can accurately and quickly obtain the semantic information of the entity in the homescene is the key to determine?
its intelligence level. In order to enhance its semantic information acquisition and knowledge reasoningability,we propose an automatic construction process of knowledge graph based on family scene. Firstly,TF-IDF algorithm is used toextract service policy keywords from text information and construct the service policy graph. Secondly, the pre - trained scenesegmentation model ( SSM) is used to identify the entities in the scene,and the scene classification model?
( SCM) is used to predict theroom category and generate structured data. Then, the family entity map is constructed according to the structured data.?
Finally, theservice policy graph and the family entity graph are merged into the family service knowledge graph and stored in the Neo4J graphdatabase.?
The experimental results show that the proposed method can automatically generate domain knowledge graph according to unstructured data. By querying the knowledge graph,the semantic information needed by service can be searched,and the service policysuitable for the current working environment can?
be generated by the robot,the robot can complete the service more intelligently,whichproves the effectiveness of the proposed method.