Cross-Lingual Few-Shot Hate Speech and Offensive Language Detection Using Meta Learning

Abstract

Automatic detection of abusive online content such as hate speech, offensive language, threats, etc. has become prevalent in social media, with multiple efforts dedicated to detecting this phenomenon in English. However, detecting hatred and abuse in low-resource languages is a non-trivial challenge. The lack of sufficient labeled data in low-resource languages and inconsistent generalization ability of transformer-based multilingual pre-trained language models for typologically diverse languages make these models inefficient in some cases. We propose a meta learning-based approach to study the problem of few-shot hate speech and offensive language detection in low-resource languages that will allow hateful or offensive content to be predicted by only observing a few labeled data items in a specific target language. We investigate the feasibility of applying a meta-learning approach in cross-lingual few-shot hate speech detection by leveraging two meta-learning models based on optimization-based and metric-based (MAML and Proto-MAML) methods. To the best of our knowledge, this is the first effort of this kind. To evaluate the performance of our approach, we consider hate speech and offensive language detection as two separate tasks and make two diverse collections of different publicly available datasets comprising 15 datasets across 8 languages for hate speech and 6 datasets across 6 languages for offensive language. Our experiments show that meta learning-based models outperform transfer learning-based models in a majority of cases, and that Proto-MAML is the best performing model, as it can quickly generalize and adapt to new languages with only a few labeled data points (generally, 16 samples per class yields an effective performance) to identify hateful or offensive content.

Hate speech detection in social media: Techniques, recent trends, and future challenges

“…Among the diverse topics investigated, the predominant data reservoir for numerous studies has been derived from Twitter. Subsequently, Facebook has been another substantial source (Bilal et al, 2022; MacAvaney et al, 2019; Mozafari et al, 2022; Rodriguez et al, 2022; Sreelakshmi et al, 2020). Additionally, alternative platforms such as YouTube comments have been utilized (Kumar Roy et al, 2022; Roy et al, 2022; Sajid et al, 2020), along with resources like Wikipedia (Beddiar et al, 2021) and diverse online platforms (Alatawi et al, 2021; Beddiar et al, 2021).…”

“…Numerous research investigations have meticulously delved into the realm of identifying hate speech, with a notable emphasis on discerning hate speech within political contexts, as is evident from the works of (Oriola & Kotze, 2020; Ribeiro et al, 2018; Wang et al, 2022). Additionally, the academic discourse extends to the scrutiny of hate speech intertwined with religious themes, as underscored by (Al‐Hassan & Al‐Dossari, 2022; Ali et al, 2021; Ghosh et al, 2023; Mozafari et al, 2022; Qureshi & Sabih, 2021; Sajid et al, 2020). Visual representation of the distribution of these topics is offered in Figure 12 and Table 9, which convey the number of instances within each thematic domain.…”

“…Race (Mathew et al, n.d.;Alatawi et al, 2021;Chiu et al, 2021;MacAvaney et al, 2019;Sajid et al, 2020;Shannaq et al, 2022;Sharma & Shrivastava, 2018;Zhang & Luo, 2018) Religion (Sajid et al, 2020;Zhang & Luo, 2018), (Mathew et al, n.d.;Ali et al, 2021;Mozafari et al, 2022;Qureshi & Sabih, 2021) Ethnicity (Ali et al, 2021;Ghosh et al, 2023;Mozafari et al, 2022;Pronoza et al, 2021;Watanabe et al, 2018) Sexist (Al-Hassan & Al-Dossari, 2022;Baydogan & Alatas, 2021;Chiu et al, 2021;MacAvaney et al, 2019;Mozafari et al, 2022;Qureshi & Sabih, 2021;Shannaq et al, 2022) Political (Govers et al, 2023;Oriola & Kotze, 2020;Plaza-Del-Arco et al, 2021;Ribeiro et al, 2017;Wang et al, 2022) Women and immigrants (Irani et al, 2021;MacAvaney et al, 2019;Zhou et al, 2020) Covid-19 (Ahmed & Lin, 2022;Khanday et al, 2022;Su et al, 2023) Gender (Ghosh et al, 2023) Crime (Mathew et al, n.d.) and those addressing hate speech within the framework of the COVID-19 outbreak, such as Khanday et al (2022) and…”

Model-Agnostic Meta-Learning for Multilingual Hate Speech Detection

“…X-MAML explores various auxiliary languages to identify the optimal composition for zero-shot cross-lingual transfer. Meta-learning has also been applied in the detection of offensive language in cross-lingual and code-mixed texts [40], [41] and other harmful content such as multilingual rumours [42]. However, the limited availability of multilingual hate speech datasets, comprising of only two or three languages, presents a challenge in finding an effective auxiliary language.…”

Hate Speech Detection for Banjarese Languages on Instagram Using Machine Learning Methods

“…They have conducted experiments for a binary hate speech classification task in Multilingual-Train Monolingual-Test, Monolingual-Train Monolingual-Test, and Language-Family-Train Monolingual Test scenarios. Mozafari et al [13] investigated the feasibility of applying a meta-learning approach in cross-lingual few-shot hate speech detection by leveraging two meta-learning models based on optimization-based and metric-based (MAML and Proto-MAML) methods. These findings demonstrate the varying performance of different machine learning approaches in hate speech detection, depending on the language and dataset under consideration.…”