Presumptive Detection of Cyberbullying on Twitter through Natural Language Processing and Machine Learning in the Spanish Language

Deep Learning-based Natural Language Processing Methods Comparison for Presumptive Detection of Cyberbullying in Social Networks

“…The label can be between 0 and 1, where the first value indicates that a comment is negative or has harassment and the second one when it is a positive or clean one. In turn, the Corpus used here has been implemented and developed by [18]. However, to establish a margin of error, we have developed a second manually classified corpus.…”

“…It should be noted that, as described in [18], a limitation of the Corpus is the impossibility of analyzing irony, since both verbs and contextual structures are not taken into consideration.…”

“…In this context, we continue the work presented in [18], where a comparison is made between two models of deep learning; the first one based on RNN+LSTM and the second one on BERT for the presumptive detection of cyberbullying in Spanish language. Our goal is to obtain accurate and efficient models in terms of execution time, using bidirectional models with memory capabilities to train the models with a Spanish language Corpus.…”

“…For this reason, works such as [17] and [18] test SVM, LR and NB models to perform sentiment analysis based on Spanish language. Antagonistically, these evaluations are focused on traditional methods for sentiment analysis tasks with NLP.…”

“…Thus, [18] presents a semi-automatic and presumptive detection system of cyberbullying; based on NB, SVM and LG techniques for the social network Twitter in Spanish language, called Spanish Cyberbullying Prevention (SPC). Thus, [18] uses sentiment analysis, machine learning and NLP to analyze phrases in Spanish from a group of trends and specific users from Ecuador, in order to identify the existence of a semantic set related to cyberbullying with up to 91% of precision, when using SVM.…”