Incm & Novasbe Pbl project - retirement law query systemincm & Novasbe Pbl project - retirement law query system: improving the performance of a bert model applied to Portuguese law by training the model-s reader with domain-specific questions and answers

• Autor(a) principal: Figueiral, Ângelo Pascoal
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/140137
• topic: Machine learning; Business analytics; Nlp; Bert; Haystack; Question answering; Bert reader training; Business analysis; Domínio/Área Científica::Ciências Sociais::Economia e Gestão

This study reflects all stages of the application of BERT for an Extractive Question Answering task applied to the domain of the Portuguese Retirement Law, from data curation through testing. Additionally, it aims to answer whether or not the effort of creating training labels and training the model is justifiable. This is assessed by measuring the performance of the trained model in comparison with that of the base model, which is only pre-trained in the Portuguese language. Since question answering accuracy improved significantly, training is deemed a crucial step to obtain state-of-the-art performance. Generally, the project showed that a BERT model is appropriate to address Extractive Question Answering on the Portuguese Retirement Law domain.