Publicação
Named Entity Recognition and Linking in a Multilingual Biomedical Setting
| Resumo: | Information analysis is an essential process for all researchers and physicians. However, the amount of biomedical literature that we currently have available and the format in which it is found make this process difficult. Therefore, it is essential to apply text mining tools to automatically obtain information from these documents. The problem is that most of these tools are not designed to deal with non-English languages, which is critical in the biomedical literature, since many of these documents are written in the authors’ native language. Although there have been organized several shared tasks where text mining tools were developed for the Spanish language, the same does not happen for the Portuguese language. However, due to the lexical similarity between the two languages, it is possible to hypothesize that the tools for the two languages may be similar and that there is an annotation transfer between Portuguese and Spanish. To contribute to the development of text mining tools for Portuguese and Spanish, this dissertation presents the ICERL (Iberian Cancer-related Entity Recognition and Linking) system, a NERL (Named Entity Recognition and Linking) system that uses deep learning and it is composed of two similar pipelines for each language, and the parallel corpus ICR (Iberian Cancer-related) corpus. Both these tools are focused on the oncology domain. The application of the ICERL system on the ICR corpus resulted in 3,999 annotations in Spanish and 3,287 in Portuguese. The similarities between the annotations of the two languages and the F1-score of 0.858 that resulted from the comparison of the Portuguese annotations with the Spanish ones confirm the hypothesis initially presented. |
|---|---|
| Autores principais: | Andrade, Vítor Daniel Torres |
| Assunto: | Literatura Biomédica Reconhecimento de Entidade Mapeamento de Entidade Aprendizagem Profunda Contexto Ibérico Teses de mestrado - 2021 |
| Ano: | 2021 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório da Universidade de Lisboa |
| Resumo: | Information analysis is an essential process for all researchers and physicians. However, the amount of biomedical literature that we currently have available and the format in which it is found make this process difficult. Therefore, it is essential to apply text mining tools to automatically obtain information from these documents. The problem is that most of these tools are not designed to deal with non-English languages, which is critical in the biomedical literature, since many of these documents are written in the authors’ native language. Although there have been organized several shared tasks where text mining tools were developed for the Spanish language, the same does not happen for the Portuguese language. However, due to the lexical similarity between the two languages, it is possible to hypothesize that the tools for the two languages may be similar and that there is an annotation transfer between Portuguese and Spanish. To contribute to the development of text mining tools for Portuguese and Spanish, this dissertation presents the ICERL (Iberian Cancer-related Entity Recognition and Linking) system, a NERL (Named Entity Recognition and Linking) system that uses deep learning and it is composed of two similar pipelines for each language, and the parallel corpus ICR (Iberian Cancer-related) corpus. Both these tools are focused on the oncology domain. The application of the ICERL system on the ICR corpus resulted in 3,999 annotations in Spanish and 3,287 in Portuguese. The similarities between the annotations of the two languages and the F1-score of 0.858 that resulted from the comparison of the Portuguese annotations with the Spanish ones confirm the hypothesis initially presented. |
|---|
Atividades financiadas
Não foram encontrados projetos financiados.