Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  

Detalhes bibliográficos
Ano de defesa: 2025
Autor(a) principal: MENDES, Karla Carollyne Pereira lattes
Orientador(a): PAUCAR CASAS, Vicente Leonardo lattes
Banca de defesa: PAUCAR CASAS, Vicente Leonardo lattes, PINTO, Mauro Sérgio Silva lattes, SILVA E SILVA, Francisco José da lattes
Tipo de documento: Dissertação
Tipo de acesso: Acesso aberto
Idioma: por
Instituição de defesa: Universidade Federal do Maranhão
Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DE ELETRICIDADE/CCET
Departamento: DEPARTAMENTO DE ENGENHARIA DA ELETRICIDADE/CCET
País: Brasil
Palavras-chave em Português:
X;
Palavras-chave em Inglês:
X;
Área do conhecimento CNPq:
Link de acesso: https://riufma.ufma.br/handle/tede/6493
Resumo: This research aims to develop a sentiment analysis framework using different classification models applied to short and informal texts, such as those commonly found on social media platforms like Twitter. The study covers traditional supervised learning algorithms, including LSTM (Long Short-Term Memory), LinearSVC, Logistic Regression, BernoulliNB, and Decision Tree, as well as more advanced models based on the Transformer architecture, with particular emphasis on RoBERTa. These models are evaluated based on their ability to accurately identify positive, negative, or neutral sentiments, considering different preprocessing, vectorization, and textual representation techniques. The proposal seeks to compare these models across various linguistic contexts and structures, highlighting their advantages and limitations within the domain of Natural Language Processing (NLP). Additionally, optimization techniques are employed to fine-tune hyperparameters, aiming to improve model performance on a dataset consisting of one million tweets. The RoBERTa and LSTM models achieved the best accuracy results, with 89.84% and 79.13% respectively, outperforming and remaining competitive with recent results in the literature.
id UFMA-2_4e67001e3351c8fcd44dd98e8e9d2e64
oai_identifier_str oai:riufma.ufma.br:tede/6493
network_acronym_str UFMA-2
network_name_str Repositório Institucional da UFMA
repository_id_str
spelling 2025-09-16T13:48:07Z2026-09-19T13:50:17Z2025-07-25MENDES, Karla Carollyne Pereira. Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa. 2025. 57 f . Dissertação( Programa de Pós-graduação em Engenharia de Eletricidade/CCET) - Universidade Federal do Maranhão, São Luís, 2025.https://riufma.ufma.br/handle/tede/6493This research aims to develop a sentiment analysis framework using different classification models applied to short and informal texts, such as those commonly found on social media platforms like Twitter. The study covers traditional supervised learning algorithms, including LSTM (Long Short-Term Memory), LinearSVC, Logistic Regression, BernoulliNB, and Decision Tree, as well as more advanced models based on the Transformer architecture, with particular emphasis on RoBERTa. These models are evaluated based on their ability to accurately identify positive, negative, or neutral sentiments, considering different preprocessing, vectorization, and textual representation techniques. The proposal seeks to compare these models across various linguistic contexts and structures, highlighting their advantages and limitations within the domain of Natural Language Processing (NLP). Additionally, optimization techniques are employed to fine-tune hyperparameters, aiming to improve model performance on a dataset consisting of one million tweets. The RoBERTa and LSTM models achieved the best accuracy results, with 89.84% and 79.13% respectively, outperforming and remaining competitive with recent results in the literature.Nesta pesquisa, desenvolve-se uma análise de sentimentos por meio da aplicação de diferentes modelos de classificação a textos curtos e informais, como aqueles encontrados em redes sociais, especificamente no Twitter, atualmente X. O estudo contempla algoritmos tradicionais de aprendizado supervisionado, como LSTM (Long Short-Term Memory), LinearSVC, Regressão Logística, BernoulliNB e Árvore de Decisão, além de modelos mais avançados baseados na arquitetura Transformer, com destaque para o modelo RoBERTa (Robustly Optimized BERT Pretraining Approach). Esses modelos são avaliados quanto à sua capacidade de identificar, com precisão, sentimentos positivos, negativos ou neutros, considerando diferentes técnicas de pré-processamento, vetorização e representação textual. A proposta busca comparar esses modelos em diferentes contextos e estruturas linguísticas, evidenciando suas vantagens e limitações no domínio do Processamento de Linguagem Natural (PLN). Adicionalmente, são empregadas técnicas de otimização para ajuste de hiperparâmetros, com o objetivo de aprimorar o desempenho dos modelos em um conjunto de dados com um milhão de tweets. Os modelos RoBERTa e LSTM foram os que tiveram melhores resultados de acurácia com 89,84% e 79,13%, que foram melhores e competitivos com os resultados da literatura recente.CAPESapplication/pdfporUniversidade Federal do MaranhãoPROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DE ELETRICIDADE/CCETUFMABrasilDEPARTAMENTO DE ENGENHARIA DA ELETRICIDADE/CCETAnálise de sentimentos;X;Processamento de Linguagem Natural;LSTM;RoBERTa;Machine Learning;Deep LearningSentiment analysis;X;Natural Language Processing;LSTM;RoBERTa;Machine Learning;Deep LearningModelos Analíticos e de SimulaçãoAnálise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  Sentiment analysis in tweets using traditional models, deep learning and RoBERTainfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/masterThesisPAUCAR CASAS, Vicente Leonardohttp://lattes.cnpq.br/1155686983267102PAUCAR CASAS, Vicente Leonardohttp://lattes.cnpq.br/1155686983267102PINTO, Mauro Sérgio Silvahttp://lattes.cnpq.br/0804349408744542SILVA E SILVA, Francisco José dahttp://lattes.cnpq.br/0770343284012942http://lattes.cnpq.br/4901737091968183MENDES, Karla Carollyne Pereirainfo:eu-repo/semantics/openAccessreponame:Repositório Institucional da UFMAinstname:Universidade Federal do Maranhão (UFMA)instacron:UFMALICENSElicense.txttext/plain; charset=utf-82393https://riufma.ufma.br/bitstreams/21f530c2-b2eb-4ca7-bc15-a390c18c9a68/downloadb8cec2d454e25b568428ee63784602fdMD55falseAnonymousREADORIGINALKARLA CAROLLYNE PEREIRA MENDES.pdfDissertação de Mestradoapplication/pdf1885697https://riufma.ufma.br/bitstreams/9e8755b0-7a8f-4d97-b0a8-c72c9a884ead/download60801a7cb987e9de4dff295d461fe0f0MD52trueAnonymousREADTEXTKARLA CAROLLYNE PEREIRA MENDES.pdf.txtWritten by FormatFilter org.dspace.app.mediafilter.TikaTextExtractionFilter on 2026-09-24T00:01:22Z (GMT).Extracted texttext/plain102579https://riufma.ufma.br/bitstreams/0f576959-221a-44ed-8677-5cb8248ab74b/downloadd97a85679a9b322cce176987f80e6756MD53falseAnonymousREADTHUMBNAILKARLA CAROLLYNE PEREIRA MENDES.pdf.jpgWritten by FormatFilter org.dspace.app.mediafilter.PDFBoxThumbnail on 2026-09-24T00:01:22Z (GMT).Generated Thumbnailimage/jpeg3297https://riufma.ufma.br/bitstreams/c22170b7-71b0-468e-808c-6c456971c263/downloadc4cfe872c8b3da6fafb2d82fe5390acaMD54falseAnonymousREADtede/64932026-10-16T23:00:39.316665Zopen.accessoai:riufma.ufma.br:tede/6493https://riufma.ufma.brRepositório InstitucionalPUBhttps://riufma.ufma.brrepositorio@ufma.bropendoar:2026-10-16T23:00:39Repositório Institucional da UFMA - Universidade Federal do Maranhão (UFMA)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
dc.title.por.fl_str_mv Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  
dc.title.alternative.eng.fl_str_mv Sentiment analysis in tweets using traditional models, deep learning and RoBERTa
title Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  
spellingShingle Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  
MENDES, Karla Carollyne Pereira
Análise de sentimentos;
X;
Processamento de Linguagem Natural;
LSTM;
RoBERTa;
Machine Learning;
Deep Learning
Sentiment analysis;
X;
Natural Language Processing;
LSTM;
RoBERTa;
Machine Learning;
Deep Learning
Modelos Analíticos e de Simulação
title_short Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  
title_full Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  
title_fullStr Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  
title_full_unstemmed Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  
title_sort Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa  
author MENDES, Karla Carollyne Pereira
author_facet MENDES, Karla Carollyne Pereira
author_role author
dc.contributor.advisor1.fl_str_mv PAUCAR CASAS, Vicente Leonardo
dc.contributor.advisor1Lattes.fl_str_mv http://lattes.cnpq.br/1155686983267102
dc.contributor.referee1.fl_str_mv PAUCAR CASAS, Vicente Leonardo
dc.contributor.referee1Lattes.fl_str_mv http://lattes.cnpq.br/1155686983267102
dc.contributor.referee2.fl_str_mv PINTO, Mauro Sérgio Silva
dc.contributor.referee2Lattes.fl_str_mv http://lattes.cnpq.br/0804349408744542
dc.contributor.referee3.fl_str_mv SILVA E SILVA, Francisco José da
dc.contributor.referee3Lattes.fl_str_mv http://lattes.cnpq.br/0770343284012942
dc.contributor.authorLattes.fl_str_mv http://lattes.cnpq.br/4901737091968183
dc.contributor.author.fl_str_mv MENDES, Karla Carollyne Pereira
contributor_str_mv PAUCAR CASAS, Vicente Leonardo
PAUCAR CASAS, Vicente Leonardo
PINTO, Mauro Sérgio Silva
SILVA E SILVA, Francisco José da
dc.subject.por.fl_str_mv Análise de sentimentos;
X;
Processamento de Linguagem Natural;
LSTM;
RoBERTa;
Machine Learning;
Deep Learning
topic Análise de sentimentos;
X;
Processamento de Linguagem Natural;
LSTM;
RoBERTa;
Machine Learning;
Deep Learning
Sentiment analysis;
X;
Natural Language Processing;
LSTM;
RoBERTa;
Machine Learning;
Deep Learning
Modelos Analíticos e de Simulação
dc.subject.eng.fl_str_mv Sentiment analysis;
X;
Natural Language Processing;
LSTM;
RoBERTa;
Machine Learning;
Deep Learning
dc.subject.cnpq.fl_str_mv Modelos Analíticos e de Simulação
description This research aims to develop a sentiment analysis framework using different classification models applied to short and informal texts, such as those commonly found on social media platforms like Twitter. The study covers traditional supervised learning algorithms, including LSTM (Long Short-Term Memory), LinearSVC, Logistic Regression, BernoulliNB, and Decision Tree, as well as more advanced models based on the Transformer architecture, with particular emphasis on RoBERTa. These models are evaluated based on their ability to accurately identify positive, negative, or neutral sentiments, considering different preprocessing, vectorization, and textual representation techniques. The proposal seeks to compare these models across various linguistic contexts and structures, highlighting their advantages and limitations within the domain of Natural Language Processing (NLP). Additionally, optimization techniques are employed to fine-tune hyperparameters, aiming to improve model performance on a dataset consisting of one million tweets. The RoBERTa and LSTM models achieved the best accuracy results, with 89.84% and 79.13% respectively, outperforming and remaining competitive with recent results in the literature.
publishDate 2025
dc.date.accessioned.fl_str_mv 2025-09-16T13:48:07Z
2026-09-19T13:50:17Z
dc.date.issued.fl_str_mv 2025-07-25
dc.type.status.fl_str_mv info:eu-repo/semantics/publishedVersion
dc.type.driver.fl_str_mv info:eu-repo/semantics/masterThesis
format masterThesis
status_str publishedVersion
dc.identifier.citation.fl_str_mv MENDES, Karla Carollyne Pereira. Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa. 2025. 57 f . Dissertação( Programa de Pós-graduação em Engenharia de Eletricidade/CCET) - Universidade Federal do Maranhão, São Luís, 2025.
dc.identifier.uri.fl_str_mv https://riufma.ufma.br/handle/tede/6493
identifier_str_mv MENDES, Karla Carollyne Pereira. Análise de sentimentos em tweets utilizando modelos tradicionais, deep learning e RoBERTa. 2025. 57 f . Dissertação( Programa de Pós-graduação em Engenharia de Eletricidade/CCET) - Universidade Federal do Maranhão, São Luís, 2025.
url https://riufma.ufma.br/handle/tede/6493
dc.language.iso.fl_str_mv por
language por
dc.rights.driver.fl_str_mv info:eu-repo/semantics/openAccess
eu_rights_str_mv openAccess
dc.format.none.fl_str_mv application/pdf
dc.publisher.none.fl_str_mv Universidade Federal do Maranhão
dc.publisher.program.fl_str_mv PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DE ELETRICIDADE/CCET
dc.publisher.initials.fl_str_mv UFMA
dc.publisher.country.fl_str_mv Brasil
dc.publisher.department.fl_str_mv DEPARTAMENTO DE ENGENHARIA DA ELETRICIDADE/CCET
publisher.none.fl_str_mv Universidade Federal do Maranhão
dc.source.none.fl_str_mv reponame:Repositório Institucional da UFMA
instname:Universidade Federal do Maranhão (UFMA)
instacron:UFMA
instname_str Universidade Federal do Maranhão (UFMA)
instacron_str UFMA
institution UFMA
reponame_str Repositório Institucional da UFMA
collection Repositório Institucional da UFMA
bitstream.url.fl_str_mv https://riufma.ufma.br/bitstreams/21f530c2-b2eb-4ca7-bc15-a390c18c9a68/download
https://riufma.ufma.br/bitstreams/9e8755b0-7a8f-4d97-b0a8-c72c9a884ead/download
https://riufma.ufma.br/bitstreams/0f576959-221a-44ed-8677-5cb8248ab74b/download
https://riufma.ufma.br/bitstreams/c22170b7-71b0-468e-808c-6c456971c263/download
bitstream.checksum.fl_str_mv b8cec2d454e25b568428ee63784602fd
60801a7cb987e9de4dff295d461fe0f0
d97a85679a9b322cce176987f80e6756
c4cfe872c8b3da6fafb2d82fe5390aca
bitstream.checksumAlgorithm.fl_str_mv MD5
MD5
MD5
MD5
repository.name.fl_str_mv Repositório Institucional da UFMA - Universidade Federal do Maranhão (UFMA)
repository.mail.fl_str_mv repositorio@ufma.br
_version_ 1880921264100474880