A análise de sentimentos e emoções de mensagens de redes sociais envolvendo serviços de entrega de alimentos
Carregando...
Tipo
TCC
Data de publicação
2024-12-09
Periódico
Citações (Scopus)
Autores
Alves, João Pedro Rodrigues
Orientador
Menezes, Mario Olimpio de
Título da Revista
ISSN da Revista
Título de Volume
Membros da banca
Programa
Resumo
Considerando a pandemia de COVID-19, houve uma expansão significativa nos servi ços de entrega, especialmente os de alimentos, tornando-se relevante a investigação das opiniões dos clientes desses serviços. Com base nessa motivação e em trabalhos anteriores, este estudo teve como objetivo investigar técnicas de mineração de texto e Aprendizagem de Máquina (AM) na análise de sentimentos (classificação entre positivo, negativo ou neutro) e emoções (classificação entre felicidade, tristeza, raiva ou neutro) em mensagens publicadas em redes sociais relacionadas aos serviços dos aplicativos Rappi, iFood e Zé Delivery.Para alcançar esse objetivo, foram realizadas as seguintes etapas: coleta de dados em redes sociais como Twitter, Bluesky e Facebook, com rotulação manual dos textos coletados; pré-processamento dos dados utilizando
técnicas de Processamento de Linguagem Natural (PLN), como remoção de stopwords, case folding e tokenization; teste da técnica SMOTE para balanceamento das classes e possível incremento na eficácia dos resultados; e experimentos com diferentes técnicas de AM. Da literatura, foram selecionados os algoritmos Support Vector Machine (SVM), Random Forest e Naive Bayes. Os resultados obtidos incluem: um conjunto de dados
final composto por 1.728 textos rotulados; um estudo comparativo entre combinações de técnicas de PLN e AM; e um repositório publicado no GitHub, documentando as combinações mais assertivas envolvendo o SVM.
Considering the COVID-19 pandemic, there was significant growth in delivery services, particularly food delivery, making it relevant to investigate customer opinions regarding these services. Based on this context and previous studies, this research aimed to explore text mining and Machine Learning (ML) techniques in sentiment analysis (classification into positive, negative, or neutral) and emotion analysis (classification into happiness, sadness, anger, or neutral) of opinion messages published on social media about the services offered by the Rappi, iFood, and Zé Delivery applications. To achieve this goal, the following steps were undertaken: data collection from social media platforms such as Twitter, Bluesky, and Facebook, followed by manual labeling of the collected texts; data preprocessing using Natural Language Processing (NLP) techniques, including stopword removal, case folding, and tokenization; testing the SMOTE technique to balance classes and potentially enhance the effectiveness of the results; and conducting experiments with different ML techniques. From the literature, the selected ML algorithms were Support Vector Machine (SVM), Random Forest, and Naive Bayes. The results obtained include: a final dataset composed of 1,728 labeled texts; a comparative study of combinations of NLP and ML techniques; and a repository published on GitHub, documenting the most accurate combinations involving SVM.
Considering the COVID-19 pandemic, there was significant growth in delivery services, particularly food delivery, making it relevant to investigate customer opinions regarding these services. Based on this context and previous studies, this research aimed to explore text mining and Machine Learning (ML) techniques in sentiment analysis (classification into positive, negative, or neutral) and emotion analysis (classification into happiness, sadness, anger, or neutral) of opinion messages published on social media about the services offered by the Rappi, iFood, and Zé Delivery applications. To achieve this goal, the following steps were undertaken: data collection from social media platforms such as Twitter, Bluesky, and Facebook, followed by manual labeling of the collected texts; data preprocessing using Natural Language Processing (NLP) techniques, including stopword removal, case folding, and tokenization; testing the SMOTE technique to balance classes and potentially enhance the effectiveness of the results; and conducting experiments with different ML techniques. From the literature, the selected ML algorithms were Support Vector Machine (SVM), Random Forest, and Naive Bayes. The results obtained include: a final dataset composed of 1,728 labeled texts; a comparative study of combinations of NLP and ML techniques; and a repository published on GitHub, documenting the most accurate combinations involving SVM.
Descrição
Indicado para publicação.
Palavras-chave
análise de dados , aprendizagem de máquina , processamento de linguagem natural (PLN). , data analysis , machine learning , natural language processing (NLP)