Predictive modeling of online social behaviour using Dynamic Graph Neural Networks
Προβλεπτική μοντελοποίηση διαδικτυακής κοινωνικής συμπεριφοράς με χρήση Δυναμικών Νευρωνικών Δικτύων Γράφων
View/
Keywords
Dynamic Graph Neural Networks (DGNN) ; Social network analysis ; Temporal dynamics ; Multi-modal data integration ; Real-time adaptation ; Predictive modeling ; Attention mechanismsAbstract
The main focus of this dissertation is to design and benchmark a Dynamic Graph Neural Network model as one of the tools for complex behavior traces recorded in a time-variant social network. Most conventional approaches miss the incorporation of temporal dynamics and multi-modal data integration that are very important for an accurate analysis of social networks. In view of these observations, this paper proposes a DGNN model combining graph convolution with mechanisms of temporal encoding to capture the structural and temporal dependencies with effectiveness. The model becomes adaptive and interpretable due to incremental learning and attention mechanisms. With respect to empirical studies on some real-world datasets, like Twitter interactions and academic collaboration graphs, DGNN models provide superior performance in predicting user behavior, information diffusion, and trends as compared to baseline models. The results of this research show the real applications of DGNN in social media analysis, marketing, and policy-related work, among others, providing really useful insights and tools for researchers and practitioners. This work goes one step closer to the real development of dynamic social network analysis by providing a strong framework for further studies and applications.
Postgraduate Studies Programme
Πληροφοριακά Συστήματα και ΥπηρεσίεςDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα Ψηφιακών ΣυστημάτωνNumber of pages
178Language
EnglishCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα