Abstract: This article studies the evolution of opinions and interpersonal influence structures in a group of agents as they discuss a sequence of issues, each of which follows an opinion dynamics model. In this work, we propose a general opinion dynamics model and an evolution of interpersonal influence structures based on the model of reflected appraisals proposed by Friedkin. Our contributions can be summarized as follows: (i) we introduce a model of opinion dynamics and evolution of interpersonal influence structures between issues viewed as a best response cost minimization to the neighbor's actions (ii) we show that DeGroot's and Friedkin-Johnsen's models of opinion dynamics and their evolution of interpersonal influence structures are particular cases of our proposed model, and (iii) we prove the existence of an equilibrium. This work is a step towards providing a solid formulation of the evolution of opinions and interpersonal influence structures over a sequence of issues.
Introduction
Most classical models of social influence, like the DeGroot model, assume that the "weight" we give to our peers' opinions is static. However, in the real world, social power is dynamic. If your arguments consistently steer a group toward a consensus on one issue, your influence likely grows when the next issue arises. This psychological phenomenon is known as reflected appraisal.
This research introduces a generalized mathematical framework to explain how social influence—or "social power"—evolves as a group discusses a sequence of topics. By framing opinion formation as a "best-response" game where agents minimize a personal cost function, the authors prove that these shifting power structures eventually reach a stable equilibrium.
A Unified Framework for Opinion Dynamics
The authors approach opinion dynamics through the lens of game theory and optimization. Instead of simply saying agents "average" their neighbors' views, they propose that each agent minimizes a quadratic cost function. This function balances three competing desires:
- Consistency: Staying close to their own current opinion.
- Conformity: Aligning with the opinions of their neighbors.
- Stubbornness: Staying true to their original "prejudice" or initial view on the topic.
By including a "prejudice" term in the cost function, the model more accurately reflects human behavior—where individuals are influenced by others but rarely abandon their core beliefs entirely. A significant insight of this work is that the famous DeGroot and Friedkin-Johnsen models are not separate entities; they are mathematically shown to be special, restricted cases of this generalized cost-minimization model.
The Evolution of Social Power
The "evolution" happens between issues. Once a group reaches a consensus on Issue $s$, the agents update their "self-appraisal" (their social power) for Issue $s+1$. This update is based on how much control they actually exerted over the final outcome of the previous issue.
To analyze this, the authors used Matrix Theory (specifically row-stochastic and Metzler matrices) and the Perron-Frobenius Theorem to determine the consensus reached in a single issue. The model treats social power as a recursive process where your power at step $s+1$ is a function of your performance at step $s$.
Key Results and Theoretical Contributions
The paper provides several rigorous proofs and simulation-based validations:
- Existence of Equilibrium: Using the Brouwer Fixed-Point Theorem, the authors proved that for their general model, an equilibrium point $x^*$ exists within the interior of the "simplex." This means a stable state exists where everyone retains some degree of self-appraisal.
- Convergence in Simulation: In numerical simulations of a small, strongly connected network of three agents, the system reached a stable state of social power after only 25 issues, regardless of the initial starting opinions.
- Mathematical Equivalence: They provided a "Generalization of Theorem III.2," which offers a simplified mapping ($F(x)$) to calculate how self-appraisals will evolve, making the model easier for other researchers to implement.
Limitations and Future Work
While the authors proved an equilibrium exists, they have not yet proven that it is unique. In some complex network topologies, there might be multiple stable states the group could fall into. Additionally, while simulations show the system always settles down, a formal mathematical proof of convergence for the most general version of their model remains a conjecture for future work.
Finally, the model assumes agents act rationally to minimize a specific cost function. Future research could involve testing this against behavioral data from social media or controlled laboratory experiments to see if human "cost functions" actually match the quadratic form proposed here.
Summary updated