Comparative Approach of Sentiment Analysis Algorithms to Classify Social Media Information Gathering in the Spanish Language

Soria J.J.; De la Cruz G.; Molina T.; Ramos-Sandoval R. · Lecture Notes in Networks and Systems (2023) · DOI: 10.1007/978-3-031-21438-7_64

Resumen

This research provides a perspective on the large amount of information shared by millions of people, who express feelings and opinions on social networks. This investigation gathered data from the social network Twitter, employing the API v2 during the month of June of 2022, concerning topics of local economic indices in Peru, with the aim to identify the precision of the different sort algorithms from analysis of feelings (SA) for the Spanish language. Tweets collected that conformed to the Corpus Data from the study were processed using the Natural Language Processing (NLP); next, they vectorizaron with the Bag method of Words (BOW) which allowed them to construct a vocabulary of 4045 tokens clean, marking the absence (0) and it is present at (1) of the word in tweets with support of bookstores of software RStudio. Finally, it was come to compare three techniques of analysis of feelings (SA) in Machine Learning (Naïve Bayes, Support Vector Machine and K-Nearest Neighbors). The results indicate a 92.13% accuracy ratio for Naïve Bayes (NB) (F1 score = 0.90026135), 93.11% for Support Vector Machine (SVM) (F1 score = 0.90026135), and 91.71% for K-Nearest Neighbors (KNN) (F1 score = 0.90026135). The results indicate that the most optimal classification technique was the SVM with an accuracy of 95.15%, making it possible to identify the best technique for classifying feelings in Spanish, applied in a thematic environment related to topics of economic indicators (e.g., inflation, unemployment, dollar) in the Peruvian context.

Instituciones: Universidad Tecnológica del Perú