Two-Population Dynamics in a Growing Network Model

We introduce a growing network evolution model with nodal attributes. The model describes the interactions between potentially violent V and non-violent N agents who have different affinities in establishing connections within their own population versus between the populations. The model is able to generate all stable triads observed in real social systems. In the framework of rate equations theory, we employ the mean-field approximation to derive analytical expressions of the degree distribution and the local clustering coefficient for each type of nodes. Analytical derivations agree well with numerical simulation results. The assortativity of the potentially violent network qualitatively resembles the connectivity pattern in terrorist networks that was recently reported. The assortativity of the network driven by aggression shows clearly different behavior than the assortativity of the networks with connections of non-aggressive nature in agreement with recent empirical results of an online social system.

💡 Research Summary

The paper introduces a novel growing‑network model that explicitly incorporates two distinct node attributes: potentially violent (V) agents and non‑violent (N) agents. Unlike most existing network models that assume a single homogeneous population, this framework allows for different affinities when forming links within the same population versus across populations. The authors motivate the model by referring to structural balance theory and empirical observations of terrorist and online social networks, where stable triads dominate and aggressive connections tend to be assortative.

Model definition: The network starts from a seed of N₀ nodes, each labeled N with probability p_N or V with probability p_V = 1 – p_N. At each discrete time step a new node is added; its type is drawn from the same distribution. The newcomer selects, on average, m_r existing nodes as “initial contacts”. If the selected node is of the same type, a link is created with probability p_s; otherwise the link is created with probability p_d = 1 – p_s. For each initial contact, the newcomer also selects, on average, m_s of the contact’s neighbors as “secondary contacts”. These secondary links are formed without an explicit type‑dependent probability; instead the existing topology implicitly biases the choice, mimicking a “friend‑of‑a‑friend” mechanism. This two‑stage attachment combines random attachment (initial contacts) with an implicit preferential attachment (secondary contacts).

The analytical treatment uses a mean‑field (rate‑equation) approach. The degree k_i(t) of a node i evolves due to (1) the random attachment of new nodes (linear in time) and (2) preferential attachment proportional to the current degree when secondary contacts are made. Separate rate equations are written for N‑nodes (Eq. 1) and V‑nodes (Eq. 2). The terms X_{j,NN}, X_{j,VV}, X_{j,NV} denote the total number of edges of each type in the whole network. Because these quantities cannot be derived analytically, the authors introduce empirical functions g(ρ_N), h(ρ_V) and q(ρ_d) that capture the probability that a randomly chosen edge belongs to each class. If one assumes that secondary contacts follow the same probabilities as initial contacts, these functions reduce to simple products: g = p_N p_s, h = p_V p_s, q = p_d.

Integrating the rate equations yields closed‑form expressions for the time evolution of a node’s degree: k_i(N)(t) = G₃