A hybrid model for linking multiple social identities across heterogeneous online social networks

Abstract

Automated online profiling consists of the accurate identification and linking of multiple online identities across heterogeneous online social networks that correspond to the same entity in the physical world. The paper proposes a hybrid profile correlation model which relies on a diversity of techniques from different application domains, such as record linkage and data integration, image and text similarity, and machine learning. It involves distance-based comparison methods and the exploitation of information produced by a social network identification process for use as external knowledge towards searches on other social networks; thus, the remaining identification tasks for the same individual are optimized. The experimental study shows that, even with limited resources, the proposed method collects and combines accurate information effectively from different online sources in a fully-automated way. The mined knowledge then becomes a powerful toolkit to carry out social engineering and other attacks, or for profit and decision-making data mining purposes.