Combined Classification Models Applied to People Personality Identification

The popularization of social networks has considerably increased the volume of data generated from the interaction between people. Understanding this data can be useful both for companies and governments and for users. This work proposes to study how to infer the behavior of people on social networks from published comments, specifically using the Myers-Briggs Typological Indicator (MBTI) in a social network focused on discussions on behavioral issues. The analysis carried out employs Natural Language Processing (NLP) techniques, resampling of the data set and classification algorithms combined by Majority Vote. The results showed 90% efficiency of the combiner with the use of random oversampling. SVM and KNN were the best individual classifiers regardless of the resampling technique used. Although smaller compared to the best individual classifier, the combination approach shows a decrease in the misclassification for INFJ and INFP classes up to 11% and 34%, respectively.