Instance Matching in Knowledge Graphs Through Dynamic, Distributed and Affinity-Preserving Random Walk
Ali Assi, Mohamed Elati, Wajdi Dhifli · 2020
A key step in the integration of data stored across independent knowledge graphs is to match instances that refer to the same real-world object (e.g., the same person). In this paper, we propose DAP-Walk, a novel approach for instance matching that is based on a dynamic and affinity-preserving Markov random walk. Our approach takes into account the local and global information mutually calculated from an association graph of candidate pairs of co-referents. Precisely, we leverage this graph to rank each candidate pair through the stationary distribution computed from the random walk on the association graph. We provide a scalable Spark-based implementation for DAP-Walk where the ranking of nodes of candidate co-referents is obtained by aggregating the distributed ranks calculated across spark workers. Experimental results on benchmark datasets show the efficiency and scalability of DAP-Walk compared to several state-of-the-art instance matching approaches.