Research And Implementation Of Rule-based Knowledge Graph Reasoning Algorithm

In the era of big data,knowledge graph,as a structured knowledge representation,has great use value.Knowledge graphs can be divided into static knowledge graphs and temporal knowledge graphs according to whether the facts contain timestamps.However,knowledge graphs suffer from sparsity and incompleteness in practice.Through the knowledge graph reasoning technology,it is possible to solve related problems by deriving potential hidden facts based on existing knowledge graph.However,most of the existing knowledge graph reasoning methods that have achieved good results are deep learning models and lack interpretability,which affects the practical application of related methods.Symbolic rule is an interpretable and expressive form of representation,and rule-based knowledge graph reasoning technology emerges as a class of interpretable reasoning methods.Existing knowledge graph reasoning technology have limitations in many fields,and further research is urgently needed.Therefore,this thesis conducts in-depth research on the rule-based knowledge graph reasoning technology for static knowledge graph scenarios and temporal knowledge graph scenarios.In the static knowledge graph scenario,this thesis proposes a rulebased multi-reasoning path collaborative reasoning method for the problem of how to effectively use the reasoning paths corresponding to the rules.The method includes learning embedding representations of inference paths based on contrastive learning and performing interpretable reasoning based on the embedding representations of reasoning paths.Experimental results on public datasets verify the effectiveness of the proposed method.In the context of temporal knowledge graphs,this thesis proposes a rule-based reasoning method based on rules and existing temporal knowledge graph embedding models to address the issue of efficient and interpretable reasoning.The method can efficiently sample potential reasoning paths from temporal knowledge graphs with.a link prediction method based on reasoning paths is also designed for interpretable reasoning.Experiment results on public datasets show that the proposed method can effectively integrate rule information,temporal information and semantic information in the reasoning path for efficient reasoning.In addition,in the static knowledge graph scenario,this thesis designs and implements a static knowledge graph link prediction system based on the characteristics of the proposed framework and user needs,and explores the practical application of related technologies.