Classificação de sentimentos em textos de e-commerce utilizando redes neurais artificiais

• Ano de defesa: 2022
• Autor(a) principal: Figueiredo, Ernanny
• Orientador(a): Miloca, Simone Aparecida
• Banca de defesa: Miloca, Simone Aparecida , Villwock, Rosangela , Ito, Giani Carla
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual do Oeste do Paraná; Cascavel
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação
• Departamento: Centro de Ciências Exatas e Tecnológicas
• País: Brasil
• Palavras-chave em Português: Classificação de textos; Redes neurais; Aprendizagem de máquina
• Palavras-chave em Inglês: Text classification; Neural networks; Machine learning
• Área do conhecimento CNPq: Ciência da Computação
• Link de acesso: https://tede.unioeste.br/handle/tede/5924

Every day, millions of people openly share their opinions on social media and comment sites about specific topics, products, services, etc. Several segments of the business market are interested in gaining information from this medium that is relevant to their business. One type of desired information is the identification of sentiments expressed by registered users in the form of opinions, as this shows agreement or disagreement related to the topic. Manual collection of such information is often not feasible due to the large amount of text. This is where machine learning techniques come into play, allowing you to organize, manage and extract knowledge so that the user of the solution can improve their business strategy. This work proposes an approach to the text classification problem applied to sentiment analysis to identify the polarity of the text, i.e., to know whether the opinion is positive or negative. The literature indicates several tools with different classifiers can be found in the ones used in this work are those whose built models incorporate classifiers based on artificial neural networks. The models were created and their performance was evaluated for a specific set of data containing the opinions of consumers who purchased health care products, with texts written in Portuguese. The effect of the preprocessing stages of the texts on the models was also studied. The results showed that artificial neural network solutions, both multilayer and recurrent, implemented in Python, reach an efficiency level close to the best and most widespread commercial tools for this task.