Semi-supervised Multi-task Learning for Multi-label Fine-grained Sexism Classification

Abstract

Sexism, a form of oppression based on one's sex, manifests itself in numerous ways and causes enormous suffering. In view of the growing number of experiences of sexism reported online, categorizing these recollections automatically can assist the fight against sexism, as it can facilitate effective analyses by gender studies researchers and government officials involved in policy making. In this paper, we investigate the fine-grained, multi-label classification of accounts (reports) of sexism. To the best of our knowledge, we work with considerably more categories of sexism than any published work through our 23-class problem formulation. Moreover, we propose a multi-task approach for fine-grained multi-label sexism classification that leverages several supporting tasks without incurring any manual labeling cost. Unlabeled accounts of sexism are utilized through unsupervised learning to help construct our multi-task setup. We also devise objective functions that exploit label correlations in the training data explicitly. Multiple proposed methods outperform the state-of-the-art for multi-label sexism classification on a recently released dataset across five standard metrics.

UL & UM6P at SemEval-2023 Task 10: Semi-Supervised Multi-task Learning for Explainable Detection of Online Sexism

“…Online sexism is a pervasive problem that can harm women and create hostile environments. Sexism detection in social media content has become an emerging field of natural language processing and social computing (Jha and Mamidi, 2017;Karlekar and Bansal, 2018;Zhang and Luo, 2019;Parikh et al, 2019;Abburi et al, 2020;Chiril et al, 2020;Rodríguez-Sánchez et al, 2021;Sen et al, 2022). In the Automatic Misogyny Identification (AMI) shared task at IberEval and EvalIta 2018, participants were tasked with identifying instances of sexist behavior in tweets and categorizing them based on a taxonomy proposed by Anzovino et al (2018).…”

“…In the same context, Chiril et al (2020) have introduced a French sexism detection method based on BERT contextualized word embeddings complemented with both linguistic features and generalization strategies. Abburi et al (2020) have proposed a semi-supervised multi-task learning method using BERT model for multi-label fine-grained sexism classification. Another line of work uses counterfactually augmented data to improve out-of-domain generalizability for sexism and hate speech detection (Sen et al, 2022).…”

IIITH at SemEval-2022 Task 5: A comparative study of deep learning models for identifying misogynous memes

“…Apart from detection of directed hateful content from tweets, there has been work on identifying categories of sexism from personal accounts such as (Karlekar and Bansal, 2018). ) created a dataset having 23 labeled categories of sexism from sexism accounts without maintaining mutual exclusivity in the categories and proposed a multi-task approach involving three auxiliary tasks for the multilabel classification in (Abburi et al, 2020).…”

Fine-Grained Multi-label Sexism Classification Using a Semi-Supervised Multi-level Neural Approach

“…The classifier combines sentence embeddings generated using a BERT [13] model with those generated from ELMo [31] and GloVe [30] embeddings using biLSTM with attention and CNN. Abburi et al [1] explore a multitask approach for semi-supervised sexism classification that deploys three auxiliary tasks such as estimating the topic proportion distribution, predicting the cluster label and detecting an account of sexism without inflicting any manual labeling cost. They also explore the objective functions that make use of label correlations present in the training data.…”

“…To the best of our knowledge, Parikh et al [28] is the only work that explores the multi-label categorization of accounts of sexism using machine learning and considers more than three categories of sexism. It provides the largest dataset containing accounts drawn from 'Everyday Sexism Project' 1 , where experiences of sexism are shared from all over the world. The textual accounts are annotated using 23 categories of sexism formulated with the help of a social scientist.…”