Extração de informação em evoluções clínicas e integração com dados farmacogenômicos

• Autor(a) principal: Bettoni, Giovani Nícolas
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da PUC_RS
• Texto Completo: https://tede2.pucrs.br/tede2/handle/tede/10359
• topic: Reconhecimento de Entidades Nomeadas; Modelos de Linguagem; Interoperabilidade; Named Entity Recognition; Language Models; Interoperability; CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO

Information Extraction (IE) covers a number of Natural Language Processing (NLP) tasks. Named Entity Recognition (NER) is a task that seeks to identify the Named Entities of a text, such as names of people, places, and organizations, classifying them in a predefined set of categories. This dissertation intends to use NLP techniques and tools for the REN task in the Biomedical domain in Portuguese. Thus, we build a specific corpus and propose two models defined in neural networks able to process the text included in clinical evolutions: BERT and a convolutional neural network (CNN). In addition, a new mechanism has been introduced to incorporate pharmacogenomic knowledge that serves as a basis for aiding clinical decisions. The results show an improvement in the measures of the BERT model compared to CNN and demonstrate that Transformers-based models are promising for advancing the performance of information extraction methods for entities in the Pharmacologic domain in Portuguese. Recognition of Named Entities in clinical evolutions is gaining popularity for improving clinical extraction projects. This study allowed the community working with NLP, in the clinical context, to obtain a formal analysis of this task, including the most successful ways of performing it.