Efficient entity resolution based on subgraph cohesion

Entity resolution has wide applications and receives considerable attentions in literature. For entity resolution, similarity functions are often used to judge whether two data objects refer to the same real-world entity. However, the similar relations determined by many commonly used similarity functions lack transitivity. This fact results in the conflict that $$A$$A and $$B$$B refer to the same entity and $$B$$B and $$C$$C refer to the same entity, but $$A$$A and $$C$$C do not refer to the same entity. To address this problem and make the group-wise entity resolution results consistent with pairwise entity resolution, this paper models the entity resolution problem as the partition of the vertices in a weighted graph into cohesive subgraphs, which is proven to be co-NP-complete. To solve this problem, an approximate algorithm with approximation ratio bound is proposed. For performing entity resolution on a large data set efficiently, a heuristic algorithm is developed to address this problem. In order to implement the heuristic algorithm efficiently, a similarity measure compatible with many measures in common usage is presented. With such similarity measure, indices and efficient implementations for the heuristic algorithm are proposed. Extensive experiments have been performed to verify the efficiency and effectiveness of the methods in this paper.