Entity resolution (ER) - the process of identifying records that refer to the same real-world entity - pervasively exists in many application areas. Nevertheless, resolving entities is hardly ever completely accurate. In this paper, we investigate a provenance-aware framework for ER. We first propose an indexing structure that can be efficiently built for provenance storage in support of an ER process. Then a generic repairing strategy, called
(CSM), is developed to control the interaction between repairs driven by must-link and cannot-link constraints. Our experimental results show that the proposed indexing structure is efficient for capturing the provenance of ER both in time and space, which is also linearly scalable over the number of matches. Our repairing algorithms can significantly reduce human efforts in leveraging the provenance of ER for identifying erroneous matches.