Learning Influential Cognitive Links in Social Networks by a New Hybrid Model for Opinion Dynamics

Abstract

A principled approach to modeling sociocognitive networks is fundamental to understanding the network interrelations which in turn can be used in many applications such as human behavior analysis or team performance assessment. More specifically, in the opinion domain, learning the cognitive links and making a proper model for causal relationships between individuals is necessary for both analysis and control purposes. There are several mathematical models for opinion dynamics. However, few of them have been tested to be consistent with real-world data. In this article, a new hybrid model for opinion dynamics is proposed and is put to test with subjective experiments. It is imperative that a realistic model considers two cognitive facts: 1) a person tends to stick to his/her previously shaped opinion and 2) the opinion of a person is affected by others (either reinforced in a positive way or undermined negatively). This article presents a novel mathematical formulation of the proposed opinion dynamics model and proves its stability too. The new model is also extended to support multiple dimensions. In the multidimensional approach, opinions about two or more subjects are considered separately. The rationale behind this is to describe the evolution of agents’ opinions on several topics. To study how the model performs in reality, some real-world experiments are conducted and the influence matrix is learned in each case. In addition, a method is introduced to extract the parameters of the model from the experimental data. It is shown that the new model predictions, after it is trained, chase the real behaviors of participants very well and result in less error compared with the previous models.