A Machine Learning Approach for Predicting Edge Motif Counts in Temporal Graphs

Temporal Networks are directed graphs where edges have labels representing the times of interactions. A fundamental problem in Network analysis is to count connected subgraph pattern, also known as "Motif Counting". This work aims to use Machine Learning to estimate Temporal Local Motif Counts, providing the formal definition to the problem, an exact algorithm, a Temporal Edge Embedding and a Machine Learning model for its estimation. This thesis also provides an empirical analysis on temporal node embedding approaches and the edge embedding techniques widely used in order to assess their efficiency and correctness of the problem under study using real world temporal networks.