水文模型知识图谱构建与应用

Hydrological modeling knowledge graph construction and application

针对水利垂直领域缺少知识组织和结构化表达的场景案例,以及通用知识抽取模型在水利垂直领域中难以达到预期精度等问题,本文以水文模型方案推荐为例,提出了知识图谱的构建框架与流程,构建了包含模型继承发展关系、应用流域、模型精度等内容的知识模型,形成基于期刊文献类非结构化数据源的多策略知识抽取方法,以及实体对齐与知识融合的规则与方法等。对水文模型领域期刊文献进行了知识抽取、融合,构建水文模型知识图谱实例,共包含节点实体14 298个,属性实体39 133个,关系36 254条,实体识别的准确率、召回率、F1值均在90%以上。对建立的图谱实例进行了可视化表达和知识应用,结果表明:该图谱实现了基于期刊文献的水文模型知识快速采集学习、组织管理,支持模型检索推荐,提高了水文模型知识的发现与使用能力,对相似场景的水利知识图谱构建具有参考价值。

Aiming at the lack of scenario cases of knowledge organization and structured expression in the water conservancy vertical domain,and the difficulty of generic knowledge extraction models in the water conservancy vertical domain to achieve the expected accuracy and other issues,this paper takes the hydrological model scheme recommendation as an example,proposes the construction framework and process of knowledge mapping,constructs a knowledge model containing model inheritance and development relationship,application watersheds,and model accuracy,etc.,and forms a multi-strategy knowledge extraction method based on the periodical literature class of unstructured data sources,as well as rules and methods for entity alignment and knowledge fusion.The knowledge model contains model inheritance and development relationship,application basin,model accuracy,etc.A multi-strategy knowledge extraction method based on unstructured data source of journal literature is formed,as well as the rules and methods of entity alignment and knowledge fusion.Knowledge extraction and fusion of journal literature in the field of hydrological modeling are carried out to construct a hydrological model knowledge graph instance,which contains a total of 14,298 node entities,39,133 attribute entities,and 36,254 relationships,and the accuracy,recall,and F1 value of the entity recognition are all above 90%.The visual expression and knowledge application of the established graph instance were carried out,and the results show that:the graph realizes rapid acquisition and learning,organization and management of hydrological model knowledge based on journal literature,supports model retrieval and recommendation,improves the discovery and use of hydrological model knowledge,and has reference value for the construction of hydrological knowledge graph in similar scenarios.