A model for ambiguation and an algorithm for disambiguation in social networks

Abstract

A common assumption when collecting network data is that objects can be uniquely identified. However, in many scenarios objects do not have a unique label giving rise to ambiguities since the mapping between observed labels and objects is not known. In this paper we consider the ambiguity problem that emerges when objects appear with more than one label in the context of social networks. We first propose a probabilistic model to introduce ambiguity in a network by duplicating vertices and adding and removing edges. Second, we propose an simple label-free algorithm to remove ambiguities by identifying duplicate vertices based only in structural features. We evaluate the performance of the algorithm under two classical random network models. Results indicate that network structure can indeed be used to identify ambiguities, yielding very high precision when local structure is preserved.