A method for detecting and quantifying information transmission biases in complex networks

Abstract:

Social networks play a major role in the transmission of information between individuals. However, one key issue of this way of transmitting information is that it can be biased: the network structure may cause some individuals to have a disproportionate visibility, hinder the communication between members in different opinion groups and so on, leading to dangerous consequences for the democratic discourse and our society in general. Despite the ubiquity of these biases however, a unified theory able to detect which ones are present in a given network and help mitigate their effects is still missing. In this work, we propose a simple method able to isolate and quantify the information transmission biases present in any network. This method is based on an orthogonal decomposition of a graph into an average component plus a set of bias variables. We explore some mathematical properties of these variables, and propose a diagrammatic representation that captures the key properties of each bias in an intuitive way. Our method sheds light into the structural inequalities arising in widely different systems, such as political communities on Twitter, linguistic communities in multicultural cities or scientific citations networks.