Learning Dissemination Strategies for External Sources in Opinion Dynamic Models with Cognitive Biases

Abdullah Al Maruf, Luyao Niu, Bhaskar Ramasubramanian, Andrew Clark, Radha Poovendran

Introduction

Traditional models of opinion dynamics, such as the DeGroot or Friedkin-Johnsen frameworks, typically assume that individuals update their beliefs through linear weighted averages of peer opinions and external information. These models often treat external inputs as simple expected values, overlooking the inherent irrationality of human decision-making. In reality, cognitive biases—such as loss aversion, distorted probability perception, and reference dependence—significantly alter how information is processed.

This research, presented at IJCAI-23, introduces a novel opinion dynamics model grounded in prospect theory (PT) to account for these biases. The authors provide a framework for external entities, such as public health agencies or advertisers, to optimize information dissemination strategies by accounting for the non-linear ways humans process "prospects" rather than just raw data.

Methodology and Network Models

The core of the proposed framework is a prospect-theoretic opinion model where an agent's update rule includes a non-linear "prospect" term. This term incorporates value functions to model loss aversion and weighting functions to capture distorted risk perceptions. Because individual bias parameters are often unknown, the researchers utilize Gaussian Process (GP) learning to estimate agent perceptions from noisy longitudinal observations of opinion changes.

To handle the complexity of multi-channel dissemination, the authors frame strategy selection as a combinatorial optimization task. They demonstrate that the objective function is submodular, a property that enables the use of efficient greedy algorithms with provable performance guarantees. The model was evaluated on three distinct network topologies:

  • Watts-Strogatz (WS): Small-world networks representing typical social circles.
  • Barabási-Albert (BA): Scale-free networks characterized by highly connected "hubs."
  • Facebook (FB): A real-world friendship network (Amherst41) containing 2,235 nodes.
Sensitivity of steady-state deviation to model parameters Ref, c+, and c- across the three network types.

Results and Analysis

The experimental results confirm that explicitly accounting for cognitive biases leads to significantly more effective influence strategies. The PT-based approach consistently outperformed standard expectation-based baselines across all tested network types.

A critical insight from the study is the role of the reference point ($Ref$). As shown in the sensitivity analysis, the steady-state deviation of the population's opinion is highly dependent on this parameter. Furthermore, the PT strategy proved exceptionally effective at shifting populations with negative initial predispositions. In the Facebook dataset, the PT-based strategy moved approximately 12.13% of the population into strong agreement with the target opinion, whereas the baseline expectation-based strategy achieved 0% movement in similar scenarios.

Comparison of final opinion shifts (> 0.5) between Prospect Theory (PT) and Expectation-based (Exp) strategies across different initial opinion distributions.

Overall, the average distance between the final population opinion and the target was reduced by 20-30% when using optimized PT strategies. This suggests that by leveraging human tendencies to over-weight low-probability events, external sources can craft messages that are far more persuasive than simple factual averages.

Conclusion

This work bridges the gap between behavioral economics and social network analysis by providing a mathematically rigorous way to incorporate human bias into opinion dynamics. By proving the submodularity of the optimization problem and employing GP learning for parameter estimation, the authors offer a scalable solution for real-world social networks. Future research may expand this model to dynamic network structures or competitive environments involving multiple conflicting external sources.

Summary updated

Abstract

The opinions of members of a population are influenced by opinions of their peers, their own predispositions, and information from external sources via one or more information channels (e.g., news, social media). Due to individual cognitive biases, the perceptual impact of and importance assigned by agents to information on each channel can be different. In this paper, we propose a model of opinion evolution that uses prospect theory to represent perception of information from the external source along each channel. Our prospect-theoretic model reflects traits observed in humans such as loss aversion, assigning inflated (deflated) values to low (high) probability events, and evaluating outcomes relative to an individually known reference point. We consider the problem of determining information dissemination strategies for the external source to adopt in order to drive opinions of individuals towards a desired value. However, computing a strategy faces a challenge that agents' initial predispositions and functions characterizing their perceptions of information disseminated might be unknown. We overcome this challenge by using Gaussian process learning to estimate these unknown parameters. When the external source sends information over multiple channels, the problem of jointly selecting optimal dissemination strategies is in general, combinatorial. We prove that this problem is submodular, and design near-optimal dissemination algorithms. We evaluate our model on three different widely used large graphs that represent real-world social interactions. Our results indicate that the external source can effectively drive opinions towards a desired value when using prospect-theory based dissemination strategies.
BibTeX
@inproceedings{maruf2023learning,
  author = {Abdullah Al Maruf and Luyao Niu and Bhaskar Ramasubramanian and Andrew Clark and Radha Poovendran},
  title = {{Learning Dissemination Strategies for External Sources in Opinion Dynamic Models with Cognitive Biases}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2023},
  doi = {10.24963/ijcai.2023/1},
}
Figure