Semantic Fingerprint and Threshold-Adaptive MinHash-LSH for Knowledge Graph Path Clustering

Knowledge graph path clustering is crucial for knowledge semantic analysis and core path mining tasks. Traditional clustering approaches struggle to balance semantic discriminability, generalization ability, and computational efficiency when handling entity-replaced isomorphic paths. This paper proposes KGPC, a knowledge graph path clustering method based on semantic fingerprint and threshold-adaptive MinHash-LSH. We convert original paths into type-based semantic fingerprints to eliminate the interference of specific entities, and adopt parameter-adaptive LSH to achieve efficient path clustering. In addition to conventional evaluation metrics including Purity, NMI, and ARI, a Comprehensive Score (CS) is introduced to synthetically evaluate clustering quality and runtime efficiency. Experiments on FB15k-237 and WN18RR demonstrate that the proposed KGPC achieves better clustering accuracy and comprehensive performance compared with representative baseline methods. Our method effectively addresses the semantic invariant clustering problem under entity replacement and facilitates knowledge graph path pruning and semantic summarization.