Framework for Modeling and Analysis of Opinion Dynamics and Influence Operations

The paper describes a flexible framework, Social Networks and Opinion Dynamics Analysis (SNODA) tool, designed to perform agent-based modeling, analysis and visualization of opinion propagation within a large social networks. The SNODA tool creates desirable virtual social structures and scenarios. To build a virtual society, users create different types of opinion agents (regular members of opinion groups, local and regional leaders), configures agent attributes and generate social network that defines information flow between agents. The SNODA framework also provides the capability to model influence operations, which are designed to shift opinions of population in a desired direction. User can create message agents, representing either influencing individuals or mass media messages, configure their attributes, and identify the social groups or individuals to be targeted by the messages (individuals with extreme opinions, influencing group leaders or groups with specific characteristics). By running a simulation, a user can visually observe dynamics of opinion landscape and monitor overall outcome at the system level. PRIMARY TRACK Application of Social Cultural Methods, Models, and Tools (MMT) SECONDARY TRACK Analytic Methods Science and Technology (S&T) DESCRIPTION The goal of designing the SNODA, the framework for modeling opinion dynamics, was three-fold: (i) create flexible framework for agent-based modeling of opinion propagation within a large artificial society (a few thousand nodes); (ii) have the capability to arrange the agents in user defined complex network which forms information flow between the agents; (iii) create capabilities to model influence operations designed to shift opinions in a desired direction. This simulates the message distribution process by targeting individuals of interest and monitoring the overall outcome at the whole society level. Most models of opinion propagation rely on a few general hypotheses of agent behavior: (i) an agent has an opinion that varies between some bounds; (ii) an agent takes into account only opinions similar to the agent’s opinion. The simplest example of that approach is the bounded confidence model [1-3]. In this model, each agent has a threshold in addition to its opinion (sometimes interpreted as an uncertainty), and it ignores opinions beyond this threshold. Our modeling principle uses modification of the bounded confidence model described by Deffuant [4]. The model can be described as follows: the population consists of N individuals, each having a continuous opinion in the interval [-1, 1] as well as continuous uncertainty from 0 to 1. At each time step, a percent of randomly chosen pairs of connected individuals interact with and influence each other. When an individual i with opinion Oi and uncertainty Ui meets another individual j with opinion Oj and uncertainty Uj, it changes its opinion to O’i calculated as follows: ) ( ) ) ( exp( ) ) ( exp( 2 2 ' i j