Automatic Polarity Identification on Twitter Using Machine Learning

Abstract

This work presents a study of emotions to analyze the polarity of a set of data that was extracted from Twitter, detailing each of the resources in the different forms that a language has, and to be able to observe feelings such as irony, sarcasm, and happiness, among others. This research can help us classify the polarity of each one of them deeply in the corpus that deals with this research work. Experimental results conducted using different machine learning methods are presented: Support Vector Machines, Naïve Bayes, Logistic regression, KNN and Random Forest, with which a classification system based on cross-validation was implemented. All experiments were performed in Python. The results obtained are shown with two different Corpus; where the first set is made up of 10,653 tweets in total divided equally each with 3551 tweets with a positive, negative and neutral label; while the second set was handled with 10% of all the tweets contained in the database mentioned in the article, where the first set shows a polarity precision of 74.9%, having Logistic Regression as the best classifier using the classification scenario known as cross validation, while the second set shows an accuracy of 78.5%, also having Random Forest as the best classifier using Cross Validation as the best classification scenario.