Analysis of Social Networks & Media — Bluesky

The project focused on the implementation and execution of a link-prediction task in the context of the decentralized social network BlueSky. A single interaction graph was used, composed of 7,175 nodes (users) and 12,309 edges (interactions: replies, reposts and quotes).

Following a study of the graph, link prediction heuristics were analyzed — in particular Common Neighbors, Jaccard Coefficient, Adamic Adar, Preferential Attachment and Shortest Path. In addition, several GNN models were implemented, including GCNConv, SAGEConv and HGTConv, together with the LinkPredictor and MLPPredictor models.