A Social Network Analysis of Articles on Social Network Analysis

A collection of articles on the statistical modelling and inference of social networks is analysed in a network fashion. The references of these articles are used to construct a citation network data set, which is almost a directed acyclic graph because only existing articles can be cited. A mixed membership stochastic block model is then applied to this data set to soft cluster the articles. The results obtained from a Gibbs sampler give us insights into the influence and the categorisation of these articles.

💡 Research Summary

The paper presents a meta‑analysis of the social network analysis (SNA) literature by constructing an citation network from a curated set of SNA articles and applying a mixed‑membership stochastic block model (MMSBM) to uncover latent thematic structures and influence patterns. The authors first compile a list of seminal SNA papers covering three broad methodological families—generative models (e.g., small‑world, preferential attachment), exponential random graph models (ERGMs), and latent models (e.g., stochastic block models, latent space models). Using each paper’s reference list, they build a directed citation network that is essentially a directed acyclic graph (DAG) because only earlier works can be cited. The data are carefully cleaned: obsolete references are updated to current DOIs, duplicates removed, and the resulting edge list is made fully reproducible.

In the modeling stage, the authors adopt the MMSBM (Airoldi et al., 2008), which allows each node (paper) to belong to multiple latent groups (topics) with a probability vector θ, while the probability of a citation between two papers depends only on their group memberships via a block matrix β. They propose two key modifications: (1) halving the number of latent variables relative to the original MMSBM, thereby reducing posterior variance and computational burden; and (2) introducing a “topological order” parameter that captures the temporal hierarchy of papers, effectively integrating citation age into the model.

Inference is performed via a Gibbs sampler. At each iteration, the authors sample the group membership proportions for each paper and the block probabilities conditional on the current state, iterating until convergence diagnostics (multiple chains, Gelman‑Rubin statistics) indicate stability. Posterior means of θ provide a soft clustering of papers, revealing the degree to which each work contributes to multiple thematic clusters, while the posterior of β quantifies citation propensity between topics.

The empirical results demonstrate that the data‑driven clusters differ from the a‑priori classification into the three methodological families. For instance, papers on “small‑world and preferential attachment” coalesce into one latent topic, whereas works on “latent space models” and “SBM extensions” form another, highlighting cross‑cutting themes that are not captured by traditional hard clustering. Moreover, several papers exhibit high mixed membership, indicating interdisciplinary influence and serving as bridges between methodological camps. The temporal component of the model shows that older, highly cited works occupy central positions in the block matrix, while newer papers tend to attach to existing topics rather than forming entirely new ones, reflecting the cumulative nature of scientific development.

The authors discuss the broader applicability of their framework. By reducing latent dimensionality and incorporating order information, the approach can be extended to dynamic MMSBMs that allow topic proportions to evolve over time, or to multi‑modal networks that include author, institution, or keyword layers. They also note that the same pipeline could be applied to other DAG‑like networks, such as software dependency graphs, demonstrating the method’s versatility.

In conclusion, the study provides a rigorous, reproducible methodology for mapping the intellectual landscape of SNA through citation network analysis. The MMSBM‑based soft clustering uncovers nuanced thematic overlaps and quantifies influence beyond simple citation counts, offering researchers a powerful tool for literature review, trend detection, and the identification of interdisciplinary bridges within any field where citation data form a directed acyclic structure.