Graph Learning of Multifaceted Motivations for Online Engagement Prediction in Counter-party Social Networks

Social media has emerged as an essential venue to invigorate online political engagement. However, political engagement is multifaceted and impacted by both individuals' self-motivation and social influence from peers and remains challenging to model in a counter-party network. Therefore, we propose a counter-party graph representation learning model to study individuals' intrinsic and extrinsic motivations for online political engagement. Firstly, we capture users' intrinsic political interests providing self-motivation from a user-topic network. Then, we encode how users cast influence on others from the inner-/counter-party through a user-user network. With the learned embedding of intrinsic and extrinsic motivations, we model the interactions between these two facets and utilize the dependency by deep sequential model decoding. Finally, extensive experiments using Twitter data related to the 2020 U.S. presidential election and the 2019 HK protests validate the model's predictive power. This study has implications for online political engagement, political participation, and political polarization.