The Laws of Opinion Dynamics in Social Systems: Theory and Inference
Abstract
Our aim is to develop a stochastic model featuring pairwise interactions among agents, generalizing the mechanism of interactions currently studied in the context of bounded confidence opinion dynamics models and bringing it closer, in spirit, to existing graph-based interaction models. In addition to opinion-dependent social exchanges, our model aims at incorporating the inherent stochasticity in interactions, imperfect exchange of opinions as well as self-beliefs, which capture the endogenous evolution of opinions innate to each agent. By incorporating opinion-dependent interactions or exchanges, bounded confidence dynamics take a good first step towards modeling opinion evolution in social systems, but this class of models and their existing variants do not yet capture some of the inherent characteristics of opinion dynamics in social systems. We analyzed the components we found to be missing in existing models, and subsequently described ways in which our model incorporates them. First, existing models assume a deterministic and thresholded behavior of agents in considering opinions of other agents. On contrary, in real life, social interactions possess a fair degree of inherent randomness, and lack sharp thresholds in terms of interactions and overall behavior. Second, in most bodies of existing work, it is often assumed that each agent has full knowledge of the opinions of the agents it interacts with. However, in practice, opinions may not be known exactly, and there may be an associated error in estimation. This estimation error can substantially impact the process of incorporation of other agents opinions in both space and time. Third, error/noise in estimating the opinion of an agent can also directly impact the actual opinion update process. Fourth, each agent may possess its own innate self-beliefs that influence its opinion, in addition to external interactions with other agents within the social system. Fifth, not all agents that share similar opinions may interact with one other, as they may not gain the opportunity to do so. In addition, the strengths of friendships, and therefore, the extent of interaction between all agents may not be the same. Our model aims to capture the above mentioned five missing elements into a stochastic framework generalizing bounded-confidence opinion dynamics. We aim at characterizing the conditions under which these dynamics are stable, in a mathematical sense, and at analyzing the implications of this result from a sociological perspective. Overall, we expect to build a stronger connection between the two bodies of work on graph-based and bounded confidence based dynamics, in addition to providing a stochastic generalization of both.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 11, 2018
- Source ID
- W911NF1510225
Entities
People
- François Baccelli
Organizations
- Army Contracting Command
- United States Army
- University of Texas at Austin