An application of graph neural networks on topic modelling with bi-partite graphs

• Ano de defesa: 2023
• Autor(a) principal: Oliveira, Bruno Fernandes Bessa de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Aprendizado de máquina; Deep learning com grafos; Graph deep learning; Machine learning; Modelagem de tópicos; Neural networks; Redes neurais; Topic modelling
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55137/tde-03012024-174804/

Graphs are data structures proper to represent real-world objects and their relationships having been widely studied in theory and with multiple examples of applications in industries and academic research. Applying graph-based data in machine learning had a significant advance with the proposal of Graph Neural Networks (GNNs), allowing the representation of this type of data in algorithms that can retain features from the graph without the need for preprocessing stage. This master\'s dissertation presents an analysis of GNNs and proposes an application on text classification using topic modelling to create descriptive variables in bi-partite graphs.