Statistical Modeling of Social Media Posting Times

Mentor:
Lead: Ana-Maria Staicu (Statistics, NC State)

Outline:  
Original posts and retweets, which reflect both individual engagement patterns and social interaction dynamics. Posting behavior can exhibit pronounced temporal structure, including bursts of activity, self-excitation, and rapid cascades driven by retweeting. These dynamics are naturally represented using point process models, particularly Hawkes processes, which capture both baseline activity and excitation induced by prior events. Despite their relevance, most studies of social media and mental health focus primarily on textual content or aggregated posting counts, often treating posting times as incidental or secondary. However, posting times themselves carry meaningful information about user engagement, attention, and reactivity, and ignoring their temporal structure can obscure important behavioral signals. This project focuses exclusively on modeling the timing of original posts and retweets using Hawkes-type processes, leveraging social media data collected during the COVID-19 lockdown.

Objectives:
The primary objective of this project is to develop and evaluate statistical models for social media posting behavior that explicitly account for the temporal dynamics of original posts and retweets. Specifically, the project aims to: (1) Develop Hawkes process – based models that distinguish baseline posting activity from self-excitation and retweet-driven cascades. (2) Characterize differences in temporal dynamics between original posts and retweets, including excitation strength, decay patterns, and burstiness. (3) Investigate how posting-time dynamics evolve over time, capturing changes in engagement intensity and reactivity over time. (4) Assess the identifiability, estimation, and inferential properties of the proposed models under realistic data conditions.

Outcomes:
The project will result in a manuscript that presents new or adapted Hawkes process-based models for analyzing social media posting times, with a focus on distinguishing original posting behavior from retweet dynamics. Simulation studies based on synthetic point process data will be used to evaluate model performance, estimation accuracy, and sensitivity to modeling assumptions. The proposed methods will be illustrated using social media data collected during the COVID-19 lockdown, highlighting the inferential value of event-time information alone. In addition, software will be developed to implement the proposed methods, enabling reproducible analysis and facilitating broader adoption of temporal point process approaches for social media data.

References:
1.     Rizoiu, M.A., Lee, Y., Mishra, S. and Xie, L., 2017. Hawkes processes for events in social media. In Frontiers of multimedia research (pp. 191-218).

 2.     Lima, R., 2023. Hawkes processes modeling, inference, and control: An overview. SIAM Review, 65(2), pp.331-374.