Estudo comparativo de modelos de classificação textual aplicados na classificação de Fake News
The present work aims to analyze the performance of three text classification models for identifying fake news. A news classification system was developed using variations of the BERT model. The models used were: BERT, DistilBERT and BERTimbau. The defined scenario was to analyze 7200 samples of...
| Autor principal: | Gusmão, Felipe dos Santos |
|---|---|
| Grau: | Trabalho de Conclusão de Curso |
| Idioma: | por |
| Publicado em: |
Brasil
2023
|
| Assuntos: | |
| Acesso em linha: |
http://riu.ufam.edu.br/handle/prefix/6934 |
| Resumo: |
|---|
|
The present work aims to analyze the performance of three text classification
models for identifying fake news. A news classification system was developed
using variations of the BERT model. The models used were: BERT, DistilBERT
and BERTimbau. The defined scenario was to analyze 7200 samples of news
in Portuguese that are pre-classified in the Fake.br corpus into 2 classes, true
news and fake news, with 3600 samples for each class. The performance of the
3 models for classifying this corpus was compared using metrics of precision,
accuracy, and F1 of each of the models. As expected, as it is a pre-trained
model in portuguese, the BERTimbau model presented the best results within
the evaluated metrics, getting 98% precision on the second experiment. |