Improving Prediction of Polarity in Tourism Domain using Convolutional Neural Network

Abstract

The analysis of the polarity of various types of comments has been enhanced by the development of Web 2.0, where millions of user-generated opinions on different sites now offer a wealth of information. Opinion mining focuses on automatically determining the polarity of posts for research and the development of real-world applications. This article aims to determine which of the proposed algorithms (Decision Trees, Support Vector Machine, Naïve Bayes, and Convolutional Neural Network) are most suitable for predicting the polarity of opinions in the tourism domain. For this purpose, a set of opinions (30,210) from the REST-MEX 2022 Sentiment Analysis competition, pertaining to hotels, attractions, and restaurants, is used. The experimental results obtained show that the Convolutional Neural Network (CNN) is the best classifier, achieving an accuracy of 98.83%, followed by the Support Vector Machine (SVM) and Naïve Bayes with accuracies of 71% and 70% respectively. The worst performance was observed with Decision Trees, which achieved 62% accuracy.