Structural Reconstruction of Signed Social Networks

Modeling real-world signed networks is a challenging task due to the highly dynamic microlevel growth processes involved in it. The network reconstruction is a problem in which we define a model that not only captures the patterns followed by different structural and spectral properties of real-world networks but also minimizes numerical error. In this article, we define a simple yet an effective network generation mechanism that can learn model parameters efficiently and produces simulated networks having structural properties close to a given input real-world signed social network. In the proposed model, corresponding to each node, a characteristic function is defined that controls its dynamics and its link formation process. A family of exponential functions is suited well to include aging and local growth, including triangle formation of different types of balanced and unbalanced triangles, and produce a wide range of degree distributions. In the proposed model, two layers are modeled independently, and further superimposition of layers is applied to get the final simulated signed network. Apart from that, the experimental results manifest that our proposed model, signed network structural reconstruction model (SNSRM), is able to replicate the characteristic properties of the various real-world signed networks more closely compared to the state-of-the-art models.