Modelling Information Warfare Dynamics to Counter Propaganda using a
Nonlinear Differential Equation with a PINN-based Learning Approach
Abstract
This research work presents a novel approach to combat propaganda and
misinformation in the context of information warfare through the
integration of a modified logistic differential equation, a
Physics-Informed Neural Network (PINN), and a comprehensive parameter
sensitivity analysis. The study leverages the LIAR dataset, which offers
a diverse range of labelled statements reflecting real-world political
discourse, making it an ideal foundation for capturing the complexities
of propaganda dynamics. The modified logistic equation, incorporating
parameters such as spread rate (α), decay rate (β), and growth rate (γ),
is employed to capture the intricate dynamics of information spread and
adoption. The PINN model uses domain knowledge and data to estimate the
underlying dynamics and unknown parameters of the logistic equation.
This allows for highly accurate classification and analysis of
propaganda spread. Thereafter, a parameter sensitivity analysis is
conducted under different attacker intensities, categorized as
aggressive, neutral, and non-aggressive. Varying population sizes are
considered to assess the implications of different parameter
combinations. The results reveal the significant influence of the spread
rate parameter (α), highlighting the aggressiveness of the attacker’s
propaganda dissemination in information warfare scenarios. Furthermore,
the proposed PINN-based model’s performance is evaluated against two
benchmark models: a Term Frequency-Inverse Document Frequency (TF- IDF)
model and the random forest classifier. The PINN model achieved an
impressive accuracy rate of 97.32% on the LIAR dataset, outperforming
both the TF-IDF model (93.87%) and the random forest classifier (95%).
This demonstrates the superior predictive power and effectiveness of the
PINN model in capturing the complexities of information spread in the
realm of propaganda and misinformation. This research lays down a robust
groundwork for research aimed at combating propaganda and pushing the
boundaries of information warfare. It achieves this by validating the
efficacy of the proposed model and showcasing its superiority when
compared to existing benchmark models. The comprehensive parameter
sensitivity analysis further enhances the understanding of the system’s
behavior and offers insights into the impact of varying parameter
values.