Enhancing Social Network Prediction in Graph Neural Networks Using Graph Theoretical Approaches: Social Network Analysis on The Facebook Ego Dataset

Abstract: This paper focuses on social network analysis and prediction by incorporating Graph theoretical concepts into the architecture of Graph Neural Networks (GNNs), with a focus on Facebook Ego Dataset. We propose an algorithm to leverage graph theoretical principles such as graph coloring, mutual friends, and weighted edge traversal to enhance link prediction tasks in Graph Neural Network. This approach optimizes Graph Neural Network’s performance by capturing local and global structural patterns within social networks. By using graph theoretical technique-based algorithm, the developed model aims on improved accuracy and diversity of friend recommendation in social networks. The study helps us to understand how integrating graph coloring helps in enhancing node embedding. The proposed algorithm generates improved social predictions with higher accuracy and meaningful insights. The results accentuate the importance of applying classical graph theoretical concepts with recent deep learning techniques, providing an efficient framework for social network prediction and analysis.

Keywords: Graph Neural Network, Product Recommendation System, Node Embeddings. Graph Coloring, User Product Graph.