Publicação
Neural information retrieval for biomedical question-answering
| Resumo: | At the rate that publicly available biomedical literature grows, current searching systems start to struggle to maintain an acceptable performance. This situation becomes more severe when a question is submitted in natural language format. Moved by this limitation, this dissertation has the main purpose of creating an automatic question answering system applied to the biomedical domain that returns for a given natural language question, the most relevant documents and their relevant snippets. The system was divided into three steps, the first consist in finding potentially relevant documents to the query. In the second step, a more powerful deep neural model will rank these documents, in a way that the query context and meaning is taken into consideration. Additionally, it was been proposed a novel deep neural model that is used in the final two steps of the system. Finally, the snippets that helped the deep neural model to rank the most relevant documents are also extracted. As a way of validation, the system results were compared with the results from this year’s BioASQ challenge, scoring the best result in first batch and third best on the last batch, while staying near to the top in the remaining batches. |
|---|---|
| Autores principais: | Almeida, Tiago Alexandre Melo |
| Assunto: | Neural networks Deep learning Information Retrieval Neural Information Retrieval Statistical Models |
| Ano: | 2019 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade de Aveiro |
| Idioma: | inglês |
| Origem: | RIA - Repositório Institucional da Universidade de Aveiro |
| Resumo: | At the rate that publicly available biomedical literature grows, current searching systems start to struggle to maintain an acceptable performance. This situation becomes more severe when a question is submitted in natural language format. Moved by this limitation, this dissertation has the main purpose of creating an automatic question answering system applied to the biomedical domain that returns for a given natural language question, the most relevant documents and their relevant snippets. The system was divided into three steps, the first consist in finding potentially relevant documents to the query. In the second step, a more powerful deep neural model will rank these documents, in a way that the query context and meaning is taken into consideration. Additionally, it was been proposed a novel deep neural model that is used in the final two steps of the system. Finally, the snippets that helped the deep neural model to rank the most relevant documents are also extracted. As a way of validation, the system results were compared with the results from this year’s BioASQ challenge, scoring the best result in first batch and third best on the last batch, while staying near to the top in the remaining batches. |
|---|
Atividades financiadas
Não foram encontrados projetos financiados.