Unsupervised Graph-Based Entity Resolution for Complex Entities

Entity resolution (ER) is the process of linking records that refer to same entity. Traditionally, this compares attribute values calculate similarities and then classifies pairs as referring entity or not based on these similarities. Recently developed graph-based ER approaches combine relationships between with improve linkage quality. Most only consider databases containing basic entities have static relationships, such publications in bibliographic databases. In contrast, temporal record addresses problem where can change over time. However, neither existing nor achieve high quality complex , an (such a person) its time while having different other at points article, we propose unsupervised framework aimed entities. Our provides five key contributions. First, propagate positive evidence encountered when use subsequent links by propagating changed. Second, employ negative applying link constraints restrict which candidate for linking. Third, leverage ambiguity disambiguate similar that, however, belong Fourth, adaptively exploit structure relationships. Fifth, using graph measures, refine matched clusters removing likely wrong records. We conduct extensive experiments seven real-world datasets from domains showing average our precision up 25% recall 29% compared several state-of-the-art techniques.