AFCMiner: Finding Absolute Fair Cliques From Attributed Social Networks for Responsible Computational Social Systems

Abstract

Cohesive subgraph mining on attributed social networks is attracting much attention in the realm of graph mining and analysis. Most existing studies on cohesive subgraph mining over attributed social networks neglect the fairness of attributes, which lead to difficulties in deploying responsible applications. Toward this end, this article formulates a new problem by introducing fairness into cliques model to mine the absolute fair cliques from attributed social networks. Specifically, this article adopts formal concept analysis (FCA) methodology to represent the given attributed social network, and extracts a set of special attributed equiconcepts to further return the absolute fair maximal cliques. Then, we develop an efficient absolute fair cliques detection algorithm AFCMiner for the cases of single-dimensional attributed social networks, multivalued attributed social networks, as well as multidimensional attributed social networks. Extensive experiments are conducted for demonstrating that the proposed AFCMiner algorithm can significantly reduce the time for finding absolute fair cliques with the correctness guarantee. Finally, a case study is also presented for uncovering the usefulness of our model.