Inferring user polarization in Twitter data with a simplified Graph Convolutional Network
DescriptionIt is common to attach state labels to data collected from an object of concern or a complex system. The state label helps categorize and generalize trends associated with other system or object states. This is useful when examining machinery, the weather but also psychological states and modes. This work investigates how observed human variables can be used both individually and within the social network structure to determine political polarization. A recent methodological framework of the Simple Graph Convolutional Neural Network (SGC) is applied to a dataset to determine its ability to accurately apply labels of polarized or not to individuals with connectivity observed in collected Twitter data. We demonstrate promising results supporting our new approach to inferring the correct label set for users engaged in an online social networking platform.