Leveraging social media for personal interests prediction with GNNs

Identifying personal interests is an important step in creating more personalized user experiences for applications ranging from recommendation systems to digital artifact preservation. In this paper, a novel approach that leverages the follower/following network of multiple users on social media to capture passive sensing and infer users’ particular interests based on their interactions and behavior patterns is presented. This approach uses a combination of deep learning and graph-based architectures to automatically identify areas of user interest in ancient artifacts, and prioritise their digitization based on the user interests captured, leveraging an extensive list of user social features, to enhance the effectiveness of similar methods, as well as the expressivity of graph neural networks on similar classification tasks. Experiments demonstrate that the proposed method is superior to recent approaches from the literature, showcasing its potential for real-world applications such as user profiling and recommendation systems.