Instance Matching in Knowledge Graphs through random walks and semantics

摘要: Abstract Instance Matching (IM) is the process of matching instances that refer to same real-world object (e.g., person) across different independent Knowledge Bases (KBs). This considered as a key step, for instance, in integration KBs. In this paper, we focus on problem IM KBs represented Graphs (KGs). We propose SBIGMat , novel approach based Markov random walks (RW). Our leverages both local and global information mutually calculated from pairwise similarity graph. Precisely, first build an expanded association graph consisting pairs candidates. Then, rank each candidate pair through stationary distribution computed RW semantic bipartite graph-based post-processing strategies operate obtained walk ranks optimize final assignment co-referents. provide scalable distributed implementation our top Spark framework evaluate it benchmark datasets instance track Ontology Alignment Evaluation Initiative (OAEI). The experiments show efficiency scalability compared several state-of-the-art approaches.