Comparação de técnicas para aumentar a eficácia e eficiência de análise de sentimento por meio da redução do número de features

Abstract

Sentiment Analysis is one of the most prevalent areas of Machine Learning, for being able to convert texts into classes. Though that conversion comes with its own drawbacks, among them are the high training cost and low general accuracy, which are compounded by the most common feature definition models, that generate features based on every word contained in the training datasets. Thus, this research has the goal of comparing means of reducing the number of features to considering both the increase efficiency, in the form of shorter training and prediction times, and the increase in efficacy, in the form of better parameters and results for predictions