• link prediction
• similarity based methods

Missing link prediction is a widely studied task of network analysis. It concerns the prediction of new links in a partially observed graph.
The thesis explores the problem from two sides: the mathematical analysis of some classical methods for link prediction and the application of these methods in a real world e-commerce network.
In the first part we present several similarity based indices grouped by the type of topological information they require and focusing on their mathematical formulation.
In the second part we analyze how missing link prediction algorithms previously presented perform in a network generated from a e-commerce data. In particular, such a network has a natural bipartite structure, that lead us to study link prediction on its projections.