软件导刊2026，Vol.25Issue(8)：60-71,12.DOI:10.11907/rjdk.251414
融合深度学习与知识图谱的导航迷路场景理解方法
Navigational Lost Scene Understanding Method Integrating Deep Learning and Knowledge Graph
摘要
Abstract
Navigation is an indispensable part of people's daily lives,and getting lost during the navigation process is a common challenge.To this end,a Navigation Lost Knowledge Graph(NLKG)is constructed using navigation text data feedback from travel activities,aiming to un-derstand navigation lost scenarios.Firstly,in order to achieve the usable scale of NLKG,more than 20 000 lost related texts were collected from the Weibo platform,and an artificial annotation corpus was constructed,covering over 10 000 high-quality entity annotations and more than 4 000 relationship annotations.Based on multidimensional information such as terrain,roads,and street environments,25 types of enti-ties and 20 types of relationships were defined,and the conceptual model of NLKG was established to achieve large-scale and structured knowledge expression.Secondly,a triplet joint extraction model NavLTR is proposed for navigation lost text,which comprehensively captures long-distance dependencies in sequence data,fully models complex relationships between navigation entities,automatically extracts entities and relationships related to lost experience,and constructs usable NLKG instances.Finally,based on NLKG,the importance analysis of lost factors,analysis of typical lost scenario factors,prediction of lost risk,correlation analysis of lost factors,and construction of an intelligent question answering agent system are conducted to demonstrate its application prospects and value.The experiment shows that the proposed method can effectively alleviate the scarcity problem of NLKG,providing a demonstration for integrating multi-source multimodal data to opti-mize navigation experience cognition,which is of great significance for optimizing navigation decisions and improving navigation experience.关键词
导航迷路/知识图谱/导航决策/时空信息抽取/多源知识整合Key words
navigation lost/knowledge graph/navigation decision-making/space-time information extraction/multi-source data integra-tion分类
信息技术与安全科学引用本文复制引用
陈虎,杨林,左泽均,罗明霞..融合深度学习与知识图谱的导航迷路场景理解方法[J].软件导刊,2026,25(8):60-71,12.基金项目
国家自然科学基金项目(42071383) （42071383）
