Distributed evaluation and convergence of self-appraisals in social networks

Xudong Chen, Ji Liu, Zhi Xu, Tamer Basar

Abstract

This paper presents a rigorous continuous-time analysis of self-appraisal dynamics within social networks. Building on the psychological concept of reflected appraisal, we model how individuals' self-perceived importance evolves through interpersonal feedback. We prove that in networks represented by rooted digraphs, these appraisals converge to a unique, stable equilibrium. Our findings reveal that extreme concentrations of influence are unstable, and at equilibrium, no single agent can dominate the network's self-appraisal distribution.

Introduction

In social psychology, the theory of "reflected appraisal" suggests that an individual's self-concept is largely a reflection of how they perceive others' evaluations of them. While traditional opinion dynamics models, such as the DeGroot model, focus on how groups reach consensus on external topics, they typically treat an agent's self-confidence as a static parameter. Recent extensions, like the DeGroot-Friedkin model, have begun to explore the evolution of self-confidence but have largely been restricted to discrete-time analyses or specific network topologies.

This research addresses these limitations by providing a continuous-time framework for a broad class of social networks—specifically those described by rooted digraphs. We investigate whether a group can reach a steady state of self-appraisal through distributed, local interactions.

Methodology

The proposed model defines a dynamical system where each agent $i$ maintains a self-appraisal score $x_i$. The evolution of this score is governed by a balance equation that reconciles two competing factors:

  1. Reflected Importance: The sum of influence neighbors attribute to agent $i$, weighted by those neighbors' own self-appraisals.
  2. Relative Importance: The degree to which agent $i$ values the opinions of their peers.

The system's stability is analyzed using graph-theoretic concepts, such as "root sets" (agents capable of reaching all others via directed paths), and matrix analysis involving infinitesimally stochastic structures. A key component of the proof involves demonstrating that the unit simplex—the space where the sum of all appraisals equals one—is a positive invariant set.

The Geometry of Social Balance

A central finding of this work is the characterization of the state space. For a system of $n$ agents, the valid self-appraisal scores reside on a unit simplex. The dynamics are designed such that the system naturally moves away from "dictatorial" states (where one agent has 100% self-confidence) and toward a balanced equilibrium.

Figure 1: The unit simplex for a three-agent system, illustrating the regions $P_i(\alpha)$ and the central equilibrium region $Q$. The vertices $e_i$ represent unstable 'dictator' states where one agent holds all influence.

As shown in Figure 1, the vertices of the simplex ($e_1, e_2, e_3$) act as repellers. If the network starts near a state where one person is dominant, the social feedback mechanism pushes the system toward the interior region $Q$, where influence is more distributed.

Key Insights and Results

The Power of Roots

The long-term distribution of self-appraisal is determined by the network's "roots." In a rooted digraph, only agents who serve as roots—those who can influence the entire network through directed paths—retain non-zero self-appraisal at equilibrium. "Non-root" agents, who receive information but do not influence the core, see their self-appraisal decay to zero over time.

The 50% Ceiling and Global Convergence

We prove that for almost all initial conditions, the system converges exponentially to a unique equilibrium point $x^*$. A striking mathematical property of this equilibrium is that for any root agent $i$, the steady-state appraisal satisfies $0 < x_i^* < 1/2$. This implies a "no-dictator" result: in a stable social system, the collective importance of the group always outweighs any single individual's influence.

Discussion and Limitations

While this model provides a robust mathematical foundation for self-appraisal dynamics, it assumes a static network topology. In real-world social systems, relationships and influence weights are often fluid and time-varying. Additionally, the model assumes "honest" appraisal; it does not yet account for strategic behavior where agents might manipulate their listening weights to artificially inflate their social standing.

Future research will explore these dynamics in time-varying networks and apply optimal control theory to investigate strategic influence maximization.

Summary updated

BibTeX
@article{chen2017distributed,
  author = {Xudong Chen and Ji Liu and Zhi Xu and Tamer Basar},
  title = {{Distributed evaluation and convergence of self-appraisals in social networks}},
  journal = {IEEE Trans. Automatic Control},
  year = {2017},
  doi = {10.1109/tac.2016.2554280},
}
Figure