A Unified Attribute Based Role Similarity Measure in Information Networks

Abstract

Similarity has been widely used in finding similar objects among complex and large scale information networks. Most of the current similarity comparison methods are the distance-based, link-based, neighborhood based similarity or the reference based similarity, and so on. They mainly make the similarity comparison with some kind of metrics but are lack of the measurement of the objects' semantics. The complex semantics and reactions within information networks require the metrics of wider aspects. In this paper, we integrate the attribute-based semantics with the role similarity and propose a unified attribute based role similarity measure method (UARS). It tries to resolve the deficiency of the current methods and achieve attribute based semantic similarity of objects besides the structural or the automorphism similarity of the original role based similarity. The development, computation and properties of UARS are given in detail. The experiment results are also showed to demonstrate the effectiveness and superiority of UARS.