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

This paper tackles the longstanding problem of disentangling selection (homophily) and influence in online social networks (SNS) by jointly modeling the evolution of users’ friendship ties and their content‑production behavior. Building on the actor‑oriented continuous‑time framework introduced by Snijders et al. (2007), the authors extend the co‑evolution model in four substantive ways. First, they treat posting activity as a continuous, non‑binary variable, discretizing the number of posts into quantiles so that subtle changes in behavior can be captured. Second, they incorporate a latent‑space representation of actors, thereby controlling for unobserved sources of homophily that would otherwise bias estimates of peer influence. Third, they allow the strength of both homophily and influence to depend on the current level of posting activity, introducing interaction terms that make the model behavior‑dependent. Fourth, they adopt a simulation‑based Method of Moments estimator powered by Markov Chain Monte Carlo (MCMC), which circumvents the intractable likelihood of continuous‑time Markov processes while still enabling statistical inference, goodness‑of‑fit testing, and counterfactual simulations.

The empirical application uses a unique longitudinal dataset from a large SNS, tracking roughly 1,200 university students over six months with observations every three weeks. Network ties are recorded as reciprocal friendships, and posting behavior is measured as the count of posts per interval, transformed into five quantile categories. The estimated parameters reveal three key patterns. (1) Homophily is present: users preferentially form ties with others who exhibit similar posting volumes, especially when their own activity level is low. (2) After a tie is formed, posting behavior tends to diverge rather than converge, indicating a “divergence” effect that runs counter to the classic notion of positive social influence. (3) The magnitude of both homophily and divergence varies with activity level; low‑activity users are more susceptible to both selection and subsequent divergence, whereas high‑activity users display more stable behavior. Including the latent‑space component markedly improves model fit (lower AIC/BIC) relative to a baseline model that only uses observable covariates, confirming that unobserved similarity can confound naïve estimates.

Beyond methodological contributions, the study offers practical insights for platform designers and marketers. Recommendation engines that rely solely on similarity may inadvertently accelerate divergence in content production, potentially reducing overall engagement. A more nuanced approach would weight similarity by current activity level, encouraging sustained posting among low‑activity users while preserving the high‑output of prolific contributors. Marketers could target low‑activity users with friend‑suggestion campaigns to boost network integration, whereas high‑activity users might be better served by amplification of existing content rather than new tie formation.

The authors acknowledge limitations such as the computational cost of MCMC‑MoM, the need to select the dimensionality of the latent space, and the focus on a single behavioral dimension. Future work could extend the framework to multiple concurrent behaviors (e.g., commenting, sharing), multiplex networks (e.g., likes, messages), and incorporate exogenous shocks (e.g., platform policy changes). Nonetheless, the paper demonstrates that a rigorously estimated continuous‑time co‑evolution model can reliably separate homophily from influence in observational SNS data, providing a powerful tool for both academic inquiry and real‑world platform optimization.