A Co-evolution Model of Network Structure and User Behavior in Online Social Networks: The Case of Network-Driven Content Generation

With the rapid growth of online social network sites (SNS), it has become imperative for platform owners and online marketers to investigate what drives content production on these platforms. However, previous research has found it difficult to statistically model these factors from observational data due to the inability to separately assess the effects of network formation and network influence. In this paper, we adopt and enhance an actor-oriented continuous-time model to jointly estimate the co-evolution of the users’ social network structure and their content production behavior using a Markov Chain Monte Carlo (MCMC)- based simulation approach. Specifically, we offer a method to analyze non-stationary and continuous behavior with network effects in the presence of observable and unobservable covariates, similar to what is observed in social media ecosystems. Leveraging a unique dataset from a large social network site, we apply our model to data on university students across six months to find that: 1) users tend to connect with others that have similar posting behavior, 2) however, after doing so, users tend to diverge in posting behavior, and 3) peer influences are sensitive to the strength of the posting behavior. Further, our method provides researchers and practitioners with a statistically rigorous approach to analyze network effects in observational data. These results provide insights and recommendations for SNS platforms to sustain an active and viable community.

💡 Research Summary

The paper tackles a central challenge in social‑network research: disentangling the simultaneous effects of network formation (who becomes friends) and network influence (how friends affect each other’s behavior) using only observational data. To address this, the authors adopt and extend the actor‑oriented continuous‑time model (often referred to as SAOM) originally developed by Snijders. Their extension allows for (1) continuous, non‑stationary behavioral variables (e.g., daily posting counts), (2) the inclusion of both observable covariates (demographics, academic major) and unobservable heterogeneity (latent user traits), and (3) separate parameterization of homophily (the tendency to form ties with similar others) and divergence (the tendency for behavior to become more dissimilar after a tie is formed).

Methodologically, the model treats network evolution and behavior evolution as two intertwined Markov processes that operate in continuous time. At each infinitesimal moment a user may either add or drop a tie (discrete network event) or adjust their posting intensity (continuous behavioral event). The probability of each micro‑step is governed by a rate function that incorporates the aforementioned effects. Estimation proceeds via a simulation‑based Markov Chain Monte Carlo (MCMC) algorithm: starting from initial parameter guesses, the algorithm simulates many micro‑step sequences, compares the simulated network‑behavior trajectories to the observed data, and updates the parameters to minimize the discrepancy. Convergence diagnostics (multiple chains, Gelman‑Rubin statistics) are employed to ensure reliable estimates.

The empirical application uses a unique six‑month panel from a large social‑network site, covering 1,200 university students. For each day the dataset records the friendship adjacency matrix and each user’s number of posts. The high‑frequency nature of the data enables the authors to capture fine‑grained fluctuations in posting behavior, which would be invisible in typical monthly or quarterly snapshots. In addition to basic demographics, the authors incorporate a survey‑derived latent variable representing each student’s intrinsic motivation to use the platform, thereby accounting for unobserved heterogeneity.

Results reveal three key patterns. First, the homophily parameter is positive and statistically significant, indicating that users are more likely to create friendships with peers who exhibit similar posting frequencies. This aligns with classic homophily theory and suggests that content production is a salient dimension for tie formation in this community. Second, the divergence parameter is negative and significant, meaning that after a tie is established, the posting behaviors of the two users tend to move apart rather than converge. The authors interpret this as a “competition” or “differentiation” effect: once connected, users may seek to distinguish themselves, perhaps to maintain a unique identity or to avoid redundancy in the information stream. Third, peer influence is not uniform; it varies non‑linearly with the focal user’s baseline posting level. High‑activity users experience stronger influence from their friends, whereas low‑activity users are relatively insulated. This finding implies that influence mechanisms are contingent on users’ existing engagement intensity.

From a practical standpoint, the study offers actionable insights for platform designers and marketers. Simply encouraging connections among similar users may boost short‑term activity, but without mechanisms to manage post‑tie divergence, the community could fragment or experience reduced content diversity. Targeted interventions—such as recommendation algorithms that balance similarity with complementary activity levels, or incentive schemes that reward collaborative content creation—could harness the observed homophily while mitigating excessive differentiation. Moreover, the pronounced influence on high‑activity users suggests that “micro‑influencers” (rather than macro‑level celebrities) could be leveraged for viral marketing campaigns, as their behavior is both responsive to peers and capable of cascading effects through the network.

The methodological contribution is equally important. By demonstrating that a continuous‑time, actor‑oriented framework can jointly estimate network and behavior dynamics from purely observational data, the authors provide a robust tool for scholars studying online communities, diffusion of innovations, or any setting where relational ties and individual actions co‑evolve. The approach is flexible enough to incorporate additional layers (e.g., multiple behavior types, multiplex ties) and to be combined with experimental manipulations for causal validation.

Limitations include the focus on a single demographic (university students) and a relatively short observation window, which may constrain external validity. The behavioral measure is limited to posting counts; extending the model to capture likes, shares, comments, or sentiment would enrich the analysis. Future work could explore longer panels, cross‑platform comparisons, and the integration of exogenous shocks (e.g., platform policy changes) to test the model’s predictive power under varying conditions.

In sum, the paper advances both theory and method: it uncovers a nuanced interplay between homophily, post‑tie divergence, and activity‑dependent peer influence in online content generation, and it equips researchers with a statistically rigorous framework for dissecting such co‑evolutionary processes in observational network data.