The proliferation of social networking services has facilitated the formation of echo chambers, where similar opinions are amplified, leading to increased opinion polarization. While much research has focused on the conditions that lead to opinion polarization, methods to mitigate it remain underexplored. This study investigates the potential of using adversarial attacks to reduce opinion polarization in social networks. By introducing small, strategic perturbations to the weights of network links, we conducted numerical simulations to observe the effects on opinion dynamics. Our results indicate that these perturbations can effectively mitigate opinion polarization, with the strength of the effect increasing alongside the perturbation parameter. Moreover, larger networks exhibited enhanced effectiveness in polarization mitigation. This research presents a novel approach to controlling opinion dynamics and offers insights into preventing the detrimental effects of polarization in social media environments.

README: Mitigating Opinion Polarization in Social Networks using Adversarial Attacks

Authors:

Michinori Ninomiya (Shizuoka University)
Genki Ichinose (Shizuoka University)
Katsumi Chiyomaru (Kyushu Institute of Technology)
Kazuhiro Takemoto (Kyushu Institute of Technology)

Data Description:

This dataset contains the following file:

Opinion_polarization.ipynb: Jupyter Notebook containing the code used to simulate opinion dynamics and adversarial attacks.
Fig1_neutral.csv: Data showing neutral convergence from Fig. 1 (left). Another set of simulations with the same parameters as used in the figure.
Fig1_radicalization.csv: Data showing radicalization from Fig. 1 (center). Another set of simulations with the same parameters as used in the figure.
Fig1_polarization.csv: Data showing polarization from Fig. 1 (right). Another set of simulations with the same parameters as used in the figure.
Fig2_neutral_control.csv: Intervention added to the left panel in Fig. 1.
Fig2_radicalization_control.csv: Intervention added to the center panel in Fig. 1.
Fig2_polarization_control.csv: Intervention added to the right panel in Fig. 1.
Fig5_radicalization_delayed_control.csv: Intervention added at t=1000 to the center panel in Fig. 1.
Fig5_polarization_delayed_control.csv: Intervention added at t=1000 to the right panel in Fig. 1.
Fig8_polarization_control.csv: Data of polarization control in Fig. 8.
Fig8_random_control.csv: Data of random control in Fig. 8.
Fig9_original.csv: Data of original (without control) in Fig. 9.
Fig9_polarization_control.csv: Data of polarization control in Fig. 9.
Fig9_random_control.csv: Data of random control in Fig. 9.

Usage Notes:

The dataset is intended for use in research on opinion dynamics and social network analysis. Researchers can replicate the simulations or use the data to explore further the effects of adversarial attacks on opinion polarization. The Jupyter Notebook includes detailed comments and explanations to facilitate understanding and modification of the code.

Contact Information:

For any questions or further information, please contact:

Genki Ichinose (ichinose.genki@shizuoka.ac.jp)

Methods

The data was generated using numerical simulations of opinion dynamics in a social network model implemented in the provided Jupyter Notebook (Opinion_polarization.ipynb). The model considers a binary issue where agents' opinions are influenced by their neighbors. Adversarial attacks were simulated by introducing small perturbations to the weights of network links. The effectiveness of these perturbations in mitigating opinion polarization was evaluated using metrics such as the mean absolute value and standard deviation of agents' opinions.

Mitigating opinion polarization in social networks using adversarial attacks

Data files

Abstract