Utilização de representações visuais para a análise sentimental em documentos de texto
Carregando...
Arquivos
Data
2021-03-01
Autores
Fiori, Mateus Hiramatsu
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
É notável o crescimento e popularização da utilização de conceitos e métodos de inteligência artificial, mais especificamente do aprendizado de máquina, para resolução de problemas cotidianos. Através da análise de dados históricos em combinação com um alto poder computacional é possível se obter resultados que superam a capacidade humana. O presente trabalho utiliza-se deste preceito para analisar e classificar documentos de texto de acordo com os sentimentos expressos nestes, sejam eles positivos ou negativos. Em outras palavras, desenvolveu-se um modelo capaz de determinar se uma crítica de determinado filme é positiva ou negativa. No trabalho, foram utilizados diversos tipos de algoritmos de aprendizagem, desde os mais simples, como a Regressão Logística, até modelos mais complexos como as Redes Neurais Convolucionais. A abordagem inicial foi transformar os dados textuais em representações numéricas coerentes, transformá-las em imagens e utilizá-las em diversos métodos de aprendizagem de máquina, fazendo uso de camadas convolucionais. Observou-se que a abordagem proposta possui uma performance melhor quando comparada aos algoritmos de aprendizagem sem as camadas convolucionais, que são ferramentas exclusivas para o tratamento de imagem. Isso mostra que o uso de técnicas de tratamento de imagem se mostra promissor quando o intuito é a analise de sentimentos.
It is remarkable the growth and popularization of the use of artificial intelligence’s concepts and methods, more specifically machine learning’s methods, for solving daily life problems. Through the analysis of historical data in combination with a high computational processing power it is possible to obtain results that surpass human capacity. The present work utilizes this precept to analyze and to classify text documents according to the feelings expressed on it, where these feelings could be described as positive or negative. In other words, we have developed an algorithm capable of determining whether a review of a given film is positive or negative. In the work, it is used several types of learning algorithms, from the simplest, such as Logistic Regression, to more complex models such as Convolutional Neural Networks. The initial approach was to transform the textual data into coherent numerical representations, transform them into images and use it in several machine learning methods, using convolutional layers. It was observed that the proposed approach has a better performance when compared to learning algorithms without convolutional layers, which are exclusive tools for image treatment. This shows that the use of image treatment techniques is promising when the intention is to analyze the feelings.
It is remarkable the growth and popularization of the use of artificial intelligence’s concepts and methods, more specifically machine learning’s methods, for solving daily life problems. Through the analysis of historical data in combination with a high computational processing power it is possible to obtain results that surpass human capacity. The present work utilizes this precept to analyze and to classify text documents according to the feelings expressed on it, where these feelings could be described as positive or negative. In other words, we have developed an algorithm capable of determining whether a review of a given film is positive or negative. In the work, it is used several types of learning algorithms, from the simplest, such as Logistic Regression, to more complex models such as Convolutional Neural Networks. The initial approach was to transform the textual data into coherent numerical representations, transform them into images and use it in several machine learning methods, using convolutional layers. It was observed that the proposed approach has a better performance when compared to learning algorithms without convolutional layers, which are exclusive tools for image treatment. This shows that the use of image treatment techniques is promising when the intention is to analyze the feelings.
Descrição
Palavras-chave
CNN, NLP, Redes neurais, Aprendizado de máquina, Aprendizado profundo, Neural networks, Machine learning, Deep learning