Entity Resolution with Attribute and Connection Graph

Entity resolution is a problem that arises in many information integration scenarios, which is to resolve the underlying entity that occurs in the data with the same or different surface forms. Besides the attributes information for each entity, there are also link or connection graph among the entities in most of the real world data. In this work, we propose an unsupervised entity resolution algorithm that utilizes both the attributes and connection graph of the entity. The algorithm propagates the similarity of each entity pair based on the connection graph in the similar way as in Page Rank. Instead of using even distribution for random access component in Page Rank, the attributes similarity of each entity pair is used to express that more similar entity pair in attributes should have larger Page Rank score. Experimental results on user profile matching on different social networking websites demonstrate the effectiveness of our new solution.