Faculty Mentor

Dr Ahmed Al-Taweel

Location

Russell Union Ballroom

Type of Research

Completed

Session Format

Poster Presentation

College

College of Science & Mathematics

Department

Department of Mathematical Sciences

Abstract

In this work, we develop a discrete-time susceptible-counter-infected-vaccinated-recovered (SCIVR) epidemiological framework to model the spread of mob-like information on social media. The model captures the interaction between susceptible neutral attention, infected mob-spreading narratives, counter or corrective discourse, low-visibility (vaccinated) content, and collective disengagement. COVID-19 twitter data were transformed into daily compartmental aggregates using sentiment polarity and retweet-based visibility thresholds, facilitating high-resolution modeling of information diffusion processes. The dataset comprises globally collected tweets between April and June 2021, filtered using keywords including “COVID19”, “coronavirus”, “quarantine”, “socialdistancing”, “staysafe”, “stayhome” and “lockdown”. Qualitative analysis of the SCIVR system includes derivation of the mob-free equilibrium (MFE), endemic equilibrium, and the basic reproduction number ℜ0. Numerical analyses and sensitivity analyses are conducted to examine the influence of key transmission and visibility parameters. To address time-varying transmission dynamics, we develop a hybrid SCIVR–deep learning model in which recurrent neural network architectures, including Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Bidirectional LSTM, and stacked LSTM networks, are employed to forecast time-dependent model parameters. By coupling a mechanistic SCIVR model with time-dependent parameter estimation based on deep learning, the proposed approach preserves model interpretability while allowing adaptive representation of the dynamics of evolving mobs in information distribution networks.

Program Description

.

DOI

10.20429/GS4.2026.019

Start Date

4-23-2026 2:00 PM

End Date

4-23-2026 4:00 PM

Download

Share

COinS
 
Apr 23rd, 2:00 PM Apr 23rd, 4:00 PM

Modeling Social Media Mob Dynamics Using a Hybrid SCVIR System with Time-Varying Parameters and Recurrent Neural Networks

Russell Union Ballroom

In this work, we develop a discrete-time susceptible-counter-infected-vaccinated-recovered (SCIVR) epidemiological framework to model the spread of mob-like information on social media. The model captures the interaction between susceptible neutral attention, infected mob-spreading narratives, counter or corrective discourse, low-visibility (vaccinated) content, and collective disengagement. COVID-19 twitter data were transformed into daily compartmental aggregates using sentiment polarity and retweet-based visibility thresholds, facilitating high-resolution modeling of information diffusion processes. The dataset comprises globally collected tweets between April and June 2021, filtered using keywords including “COVID19”, “coronavirus”, “quarantine”, “socialdistancing”, “staysafe”, “stayhome” and “lockdown”. Qualitative analysis of the SCIVR system includes derivation of the mob-free equilibrium (MFE), endemic equilibrium, and the basic reproduction number ℜ0. Numerical analyses and sensitivity analyses are conducted to examine the influence of key transmission and visibility parameters. To address time-varying transmission dynamics, we develop a hybrid SCIVR–deep learning model in which recurrent neural network architectures, including Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Bidirectional LSTM, and stacked LSTM networks, are employed to forecast time-dependent model parameters. By coupling a mechanistic SCIVR model with time-dependent parameter estimation based on deep learning, the proposed approach preserves model interpretability while allowing adaptive representation of the dynamics of evolving mobs in information distribution networks.