知识图谱推理的理论与方法研究

As an effective approach for describing and understanding entities and relations between entities in real world, knowledge graphs have wide application prospects in the field of semantic search, intelligent questions answering, human brain-like intelligence, intelligent city, and so on. However, as the scale of knowledge graph increase, the incompleteness and inaccuracy of query results caused by the inherent incompleteness and inconsistency of knowledge graph becomes a large obstacle to intelligent applications of knowledge graph. Therefore, this project studies the theory and method of knowledge graph inference for systematical overcome this obstacle from three aspects which include exploration of implicit knowledge, robust and intelligent query mechanism, and detection and elimination of inconsistency. Specifically, a logical closure inference rule set and distributed closure inference algorithm for knowledge graph is constructed to increase the completeness of knowledge graph. Secondly, an inference system for equivalent query derivation and method for multiple query results fusion are established to improve the completeness and accuracy of query results. Finally, a query based approach for inconsistency detecting inference and process is proposed to gradually eliminate inconsistencies of knowledge graph. This project will not only complete the theory and method of knowledge graph inference, but also provide the technical support for intelligent applications of knowledge graph.

作为描述和理解现实世界中的每一个实体以及他们之间的相互关系的有效手段，知识图谱在的语义搜索、智能问答、类人智能以及智能城市等领域有着广泛的应用前景。然而，随着知识图谱规模的扩大，知识图谱内在的不完整性和不一致性导致的查询结果的不完整性和不准确性已经成为阻碍知识图谱智能应用的难题。为此，本项目拟从知识图谱中隐含知识发掘、鲁棒智能的查询机制构建和不一致信息的检测与消除这三个角度出发，研究系统解决这一难题的知识图谱推理的理论与方法。首先，构造知识图谱逻辑闭包推理规则集合及有效的分布式闭包推理算法，提升知识图谱的完整性；其次，建立能够支持查询的等价形式推导的推理系统以及多查询结果的融合方法，提高查询结果的完整性和准确性；最后，提出基于查询的不一致性检测推理和处理方法，在使用过程中逐步消除知识图谱的不一致性。本项目的研究将进一步完善知识图谱推理的理论与方法，为知识图谱的智能应用提供技术支撑。

作为描述和理解现实世界中的每一个实体以及他们之间的相互关系的有效手段，知识图谱在的语义搜索、智能问答、类人智能以及智能城市等领域有着广泛的应用前景。然而，随着知识图谱规模的扩大，知识图谱内在的不完整性和不一致性导致的查询结果的不完整性和不准确性已经成为阻碍知识图谱智能应用的难题。针对上述问题，本项目从知识图谱理论闭包推理、基于等价查询扩展的查询结果补全和基于极小不协调子集的不一致信息的检测与消除这三个方面开展知识图谱推理的理论与方法研究。分析OWL2 RL规则集及规则间的依赖关系，设计高效知识图谱分布式闭包推理算法和增量式闭包推理算法；研究SPARQL查询的等价重写方法，提高查询结果的完整性和准确性；提出基于OWL2 RL规则集的不一致性检测推理和导致不一致的极小子集计算方法，在使用过程中逐步消除知识图谱的不一致性。项目执行期间在国际上发表了一系列原创成果，并在苏州进行项目成果的应用转化工作。

内容获取失败，请点击重试