Exploring Information Acquisition in Social Learning

Yiqing Lin, Zhanjiang Chen, Huisheng Wang, H. Vicky Zhao

Abstract

Social learning, a fundamental process through which individuals shape their beliefs and perspectives via observation and interaction with others, is critical for the development of our society and the functioning of social governance. Prior works on social learning usually assume that the initial beliefs are given and focus on the update rule. With the recent proliferation of online social networks, there is an avalanche amount of information, which may significantly influence users' initial beliefs. In this paper, we use the rational inattention theory to model how agents acquire information to form initial beliefs and assess its influence on their adjustments in beliefs. Furthermore, we analyze the dynamic evolution of belief distribution among agents. Simulations and social experiments are conducted to validate our proposed model and analyze the impact of model parameters on belief dynamics.

Introduction

Traditional social learning models, such as Bayesian learning, typically concentrate on the update rules—how individuals adjust their opinions after interacting with peers. However, these models often overlook the critical "acquisition phase": the process by which individuals form their initial beliefs from the vast "information avalanches" characteristic of modern social media. This paper addresses this gap by applying Rational Inattention (RI) theory to model how limited human attention affects initial belief formation and how this "attentional investment" subsequently dictates social influence.

Methodology: The Two-Stage Framework

The researchers propose a framework divided into two distinct stages: Information Acquisition and Belief Update.

The two-stage system model: Stage I (Information Acquisition) and Stage II (Belief Update).

In Stage I (Information Acquisition), agents act as rational actors who trade off the cognitive cost of processing information against the utility of accuracy. Using a power-law cost function, $C(x) = ax^{-b}$, the model represents the effort required to reduce uncertainty (variance) in an initial belief.

In Stage II (Belief Update), these agents enter a social network (modeled as a Barabási-Albert scale-free network). Their willingness to be influenced by social signals is inversely proportional to the effort they invested in Stage I. An agent who "paid" a high cognitive cost to form a precise initial belief will be more "stubborn" and less swayed by neighbors.

Experimental Validation and Results

To validate the model, a social experiment was conducted with 137 subjects tasked with estimating proportions in images under varying time ($T$) and reward ($M$) constraints.

Experimental validation of the cost function (left) and simulation of belief convergence (right).

As shown in Figure 3, the RI model accurately predicted the variance of human reports with a low average relative error of 6.63%. The results confirm that increasing rewards ($M$) or available time ($T$) leads to lower-variance initial signals, as predicted by the power-law cost function.

Furthermore, the simulations demonstrate that while agents may start with widely varying initial beliefs based on their individual attentional effort, the social network eventually facilitates convergence toward the true system state ($\theta$). The network acts as a "safety net," pulling even the most "lazy" or inaccurate agents toward the truth over time, though their path to convergence is more heavily influenced by their social connections.

Conclusion

By integrating Rational Inattention into social learning theory, this work provides a more realistic model of human belief dynamics in the information age. It highlights that our susceptibility to social influence is not just a personality trait, but a consequence of the cognitive effort we invest in forming our initial opinions. Future research will incorporate prior experiences and individual differences in cognitive costs to further refine these insights.

Summary updated

Abstract

Social learning, a fundamental process through which individuals shape their beliefs and perspectives via observation and interaction with others, is critical for the development of our society and the functioning of social governance. Prior works on social learning usually assume that the initial beliefs are given and focus on the update rule. With the recent proliferation of online social networks, there is an avalanche amount of information, which may significantly influence users' initial beliefs. In this paper, we use the rational inattention theory to model how agents acquire information to form initial beliefs and assess its influence on their adjustments in beliefs. Furthermore, we analyze the dynamic evolution of belief distribution among agents. Simulations and social experiments are conducted to validate our proposed model and analyze the impact of model parameters on belief dynamics.
BibTeX
@misc{lin2023exploring,
  author = {Yiqing Lin and Zhanjiang Chen and Huisheng Wang and H. Vicky Zhao},
  title = {{Exploring Information Acquisition in Social Learning}},
  year = {2023},
  eprint = {2309.05854},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2309.05854},
}
Figure