Συγκριτική μελέτη αλγορίθμων κοινωνικής σύστασης μέσω νευρωνικών δικτύων γράφων
A Comparative study of social recommendation algorithms via graph neural networks
View/
Keywords
Νευρωνικά δίκτυα γράφων ; GNN ; CGN ; RecSYS ; Graph neural networks ; Social recommendationAbstract
This postgraduate dissertation is a comparative study of social recommendation algorithms via graph neural networks. In recent years, the importance of graph NN in solving machine learning problems have grown and a large amount of this type of architectures has been generated yearly with applications in social networks, and e-commerce. There are many classical implementations on matrix completion and collaborative filtering methods, and in recent years graphs representing the complex structure of modern databases have been used for inference. Problems arise from feature architecting and latent object embeddings, which become more prominent by the lack of topology and permutation variance of said structures. There is a push towards end to end training and use of state of the art neural models such as GAT, CGN etc. In this paper some of the best current implementations of these social recommendation architectures are tested and compared together, using common opensource datasets, that contain users items and interactions in graph form.