地理知识图谱实体关联关系建模与补全方法研究

Geographic Knowledge Graph is the basis for solving the problem of cross-domain, multi-source and heterogeneous generalized spatial data interoperability. It can extend the geographic information service from the traditional "space" and "time" dimensions to new "activities" and "relations" dimension. That is of great significance to realize the fusion and intelligent analysis of multi-source data, information and intelligence. .At present, the number of Knowledge Graphs is increasing and the scale is expanding constantly. However, lots of Knowledge Graphs are not complete. Most existing methods of Knowledge Graph completion are aimed at static data. And, lack of consideration on Temporal-spatial information, results in the completion and reasoning of geographic Knowledge Graph imperfect..With respect to the problems mentioned above, this study carries out a research on Geo-Knowledge Graph completion on the basis of the Geo-Knowledge Graph construction. We focus on the following aspects: (1) methodology of Global geo-ontology library dynamic construction, (2) Geo-entity correlation model construction based on the correlative route, (3) quantification of geographic entity correlation based on the Logistics regression model, (4) temporal link prediction algorithm based on the recursive neural network. All the research achievements will be applied on the prototype system to validate and evaluate the key technologies proposed in this study. Through this research, the completeness and accuracy of the Geo-Knowledge Graph will be improved, which is of great significance for completing the theory of Geo-Knowledge Graph and promoting its practical application.

地理知识图谱是解决跨域、多源、异构的广义空间数据互联互通问题的基础，能够将地理信息服务从传统的“空间”和“时间”维度延伸到“活动”和“关系”的新维度，对于实现多源数据、信息与情报的融合和智能分析具有重要意义。目前知识图谱数量不断增多、规模不断扩大，但是大量知识图谱并不完整，而当前已有的知识图谱补全方法大多针对静态数据，对时空信息考虑不足，从而导致地理知识图谱补全和推理结果不完善。针对这一问题，本项目在实现地理知识图谱构建基础上进行深入拓展，开展地理知识图谱补全研究，重点包括：全局地理本体库动态构建方法，基于关联路径的地理实体关联模型构建，基于回归模型的地理实体关联关系量化，基于递归神经网络的地理实体时序链接预测算法，最终构建原型系统进行实践，对本项目关键技术方法进行验证评价。通过本项目研究，能够提升地理知识图谱的完备性和准确率，对于完善地理知识图谱理论、推动实际应用具有重要意义。

知识图谱在地理信息科学领域，已经与地理信息科学方法论和分析模型进行了深度融合，在智慧出行、智慧城市、旅游推荐、位置服务和防灾减灾等方面有了显著的进展。目前地理知识图谱数量不断增多、规模不断扩大，但是大量知识图谱并不完整，而当前已有的知识图谱补全方法大多针对静态数据，对时空信息考虑不足，从而导致地理知识图谱补全和推理结果不完善。本项目面向地理知识图谱实体关系建模，进一步扩展面向时空事件、领域数据（综合兵要数据）的知识图谱构建与补全的理论方法及其应用进行深入研究。本项目在实现地理知识图谱构建基础上进行深入拓展，开展地理知识图谱补全研究，重点包括：地理本体库构建和扩展方法，基于关联路径的地理实体关联模型构建，基于回归模型的地理实体关联关系量化，最终构建原型系统进行实践，对本项目关键技术方法进行验证评价。通过本项目研究，能够提升地理知识图谱的完备性和准确率，对于完善地理知识图谱理论、推动实际应用具有重要意义。.本项目的研究工作进展顺利，已达到预期的研究目标，在本体扩展取得了良好的研究成果，进一步为地理知识图谱应用的研究工作奠定了基础，申请了河南省科技攻关项目，并顺利结题。研究成果在国内外重要学术会议和期刊上发表论文22篇，SCI检索1篇，EI检索6篇；申请软件著作权2个；申请并被受理国家发明专利1个；培养了1名博士生，3名硕士生。本项目建立了地理空间情报知识图谱构建系统，应用于军事领域知识图谱构建，取得了良好的应用效果。