Study of different models for sentiment analysis and language representation

Abstract

[en] In this project we study three different models of artificial intelligence to carry out the process of sentiment analysis, which consists of determining the polarity of a text; that is, detecting whether it is positive, negative or neutral. The first model studied is a neural network, specifically a long short-term memory, which uses deep learning techniques. We delve deeper into the study of its structure and operation, unmasking all the mathematics behind it. The other two models belong to machine learning: logistic regression and Naive Bayes. We emphasize the study of its parameters and optimization, with the intention to understand the learning process of each one. Finally, we apply the results and techniques developed to implement a Python program with each model in order to detect the sentiment of thousands of reviews from social media of different bars and restaurants. We dedicate a whole chapter to give the results, the analysis of each one and a comparison between them.