Abstract:Abstract Homophily is an important social tendency that has a profound impact on the formation of social connections and the diffusion of information between individuals physically or online. In analogy to the known sociological concept of homophily, we propose a new complex networks concept named Group Homophily, that measures the preference of an entire group of nodes to interact with members of its own type, rather than with members of other groups (self-liking). The proposed Group Homophily framework is based on entropy considerations and is a general concept that could be useful in many different types of complex networks where nodes are not homogeneous but rather belong to different classes. We demonstrate the usefulness of Group Homophily on a vocational network where workers change jobs between companies. A network is constructed from the career paths (CV) of over 2:1 million workers, that move ∼ 8 million times between ∼ 2:5 million companies during two decades. In this network, companies are categorised by two methods: (1) their industrial sectors which is based upon their business occupation. (2) Companies are also categorised by their interactions with other companies through their exchange of workers and the community classes that this workers’ exchanges forms. Group Homophily estimates the entropy in the fit between these two classification methods. We find for example, that the Healthcare and Energy sectors followed by Financial and IT sectors, are sectors with high Group Homophily, i.e. have stronger self-liking for workers with previous experience in their own sector. In contrast, Industrials is a sector that show weaker self-liking and can easily exchange workers that have no previous experience in its own sector. Although the Energy sector has high levels of Group Homophily, we find that this is mainly true for educated individuals. In the Healthcare sector in contrast, this preference is true for both high and low educated individuals. Also, we find that managers generally face lower levels of group homophily and can therefore change their sectors of occupation with greater ease compared to non-managers. While we demonstrate the framework of Group Homophily in the context of vocational networks where workers change jobs between companies, Group Homophily is a very general measure that can be practically applied in almost any type of complex network. For example, it can measure the self-liking (disciplinary versus interdisciplinary) levels of different scientific domains, or to measure the tendencies of scientific collaboration in different countries or universities. It is a general new measure that estimates the self-liking between groups of nodes in a network with heterogeneous nodes types.

A novel method for assessing and measuring homophily in networks through second-order statistics

Network homophily via tail inequalities

Conformity: A Path-Aware Homophily Measure for Node-Attributed Networks

Unveiling homophily beyond the pool of opportunities

Homophily Within and Across Groups

The Homophily Principle in Social Network Analysis

Homophily as a Process Generating Social Networks: Insights from Social Distance Attachment Model

Revisiting Graph Homophily Measures

The Paradox of Second-Order Homophily in Networks

What Is Missing In Homophily? Disentangling Graph Homophily For Graph Neural Networks

Homophily Preserving Community Detection.

Group Homophily as a Measure of Self-Liking of Communities: Application in Vocational Networks

Second-Order Centrality Correlation in Scale-Free Networks

Second-Order Markov Assumption Based Bayes Classifier for Networked Data with Heterophily

Statistically validated networks in bipartite complex systems

A primer on statistically validated networks

Classification in Networked Data with Heterophily

Multidimensional homophily

A comparison between the Coleman homophily index and BA-homophily metric with a random network of unequal group sizes

Homophyly/Kinship Hypothesis: Natural Communities, and Predicting in Networks

On the inadequacy of nominal assortativity for assessing homophily in networks