Classificação de gêneros e faixas etárias em redes sociais online por meio de técnicas de aprendizagem multidimensional

Abstract

Due to the large volume of content generated by users on Online Social Networks (OSN), organizations have applied sentiment analysis or opinion mining techniques to obtain information about people or entities of interest. An entity can be products, services, people, governmental and non-governmental institutions, public policies, among other types. The classification of genders and age groups supports the analysis of sentiment and opinion, as they help to obtain a more precise feeling or opinion. However, information about gender and age-group may be hidden or incorrectly filled out in the OSN. In the literature, several approaches are used in order to classify genders and age groups. However, in this work, a new set of features is used to classify genders, and age groups, through multidimensional learning. Thus, the main objective of this work is to develop a new model of classification of genders and age groups with data extracted from OSN Twitter, using the transformation methods Classifier Chains (CC) and Label Powerset (LP), and through machine learning techniques based on rules, linear algebra, and probability. This study works with a new database containing 8000 instances extracted from Twitter. The best subsets of user profile features are evaluated, as well as multidimensional learning models using different performance metrics. Through the experiments, a multidimensional classification model was obtained in the test phase, with 0.999 of F1 micro-average for genders and 0.923 for age groups. The results of the classification of genders surpassed most of the related works, and the performance of the classification of age groups is quite competitive.