Research On Graph Aggregation In Knowledge Graph

Knowledge Graph provides a powerful primitive for modeling relationships among entities in a variety of domains.Recently,with the explosive growth of datasets,data transmission,storage,and Graph Mining on Knowledge Graph(GMKG)have to rise to the challenge.The graph aggregation method provides a solution to these problems.Graph aggregation is a process of effectively merging nodes and edges while maintaining the overall structure of the graph,thereby obtaining a more concise and abstract hypergraph.Being different from the traditional graph aggregation method,a lossless graph aggregation method for knowledge graph is proposed in this thesis.It can effectively divide a knowledge graph into a small-scale summarized knowledge graph containing significant knowledge,and a detailed graph consisting of other subordinate knowledge.The separation of summary and detailed knowledge in a knowledge graph will expedite GMKG tasks by greatly reducing the search space,and the detailed knowledge can be re-added to the summary knowledge to restore the original knowledge graph.The graph aggregation method proposed in this thesis contains multiple graph aggregation strategies which can effectively identify and compress specific structures in the knowledge graph,thereby improving the efficiency of knowledge graph mining algorithms for these specific structures.The main work of this thesis includes:(1)A focused aggregation method is proposed,including equivalent aggregation,dependent aggregation and graph aggregation strategies.These aggregation strategies respectively mine the equivalent structure,dependent structure and frequent pattern structure in the dataset.(2)A non-focused aggregation method is proposed,which can remove irrelevant entities according to the user's needs and return an summarized graph containing only specific types of entities.(3)A set of experiments is conducted to measure the compression performance and the shortest path search performance.A user-friendly interface is provided to show the aggregated results.The method proposed in this thesis provides a new idea of the aggregation method for assisting the mining tasks,which is of great significance for the research and practice of the graph aggregation method.