Automatic Identification of Misogynistic Content on Social Networks: An Approach based on Knowledge Transfer from Songs

Ricardo Calderón-Suarez, Rosa María Ortega-Mendoza, Marco Antonio Márquez-Vera, Félix Agustín Castro-Espinoza


This research paper presents a summary ofthe thesis “Automatic Detection of Misogynistic Contentin Social Networks through Knowledge Transfer fromSongs”, where the main idea is to leverage the existingknowledge of some songs to transfer linguistic patternsthat help to identify manifestations of misogyny insocial media. In particular, several learning transfertechniques were analyzed. In addition, a methodologyis presented to build, automatically, a collection ofsongs and another of phrases, both with instanceslabeled according to the presence or absence ofmisogynistic content. The major contribution of thisresearch is a data augmentation method that increasesthe generalization capability of the misogyny detectionmodels by transferring the semantic richness containedin song lyrics. The proposed approach was evaluatedin benchmark collections containing texts in Spanishand English, obtaining encouraging results. Comparedto robust state-of-the-art approaches, the proposedapproach obtained competitive results in English andsignificant gains in Spanish. This research confirmedthe existence of valuable linguistic knowledge in songs,which can be transferred to detect misogynistic contentin social media.


Transfer learning, data augmentation, mysogyny detection, social media

Full Text: PDF (Spanish)