A bilingual Malay-English social media dataset for binary hate speech detection

Area of Science:

• Natural Language Processing (NLP)
• Computational Linguistics
• Social Media Analysis

Background:

• Online hate speech is a significant threat to user safety and social cohesion.
• Existing hate speech datasets are often monolingual, neglecting underrepresented languages like Malay.
• Southeast Asian languages lack sufficient resources for NLP research, hindering hate speech detection.

Purpose of the Study:

• To address the gap in multilingual hate speech datasets by creating a balanced, quality-controlled resource.
• To support machine learning applications for hate speech detection in low-resource, multilingual settings.
• To facilitate binary classification tasks for practical, early-stage hate speech detection systems.

Main Methods:

• Curated 26,985 bilingual Malay-English social media texts from five public sources.
• Combined human annotation with controlled pseudo-labeling for high-confidence data.
• Ensured data quality through rigorous filtering and validation processes.

Main Results:

• Developed a balanced dataset with 13,609 English and 13,376 Malay texts.
• Provided data in UTF-8 encoded CSV format with clear labels and metadata.
• Dataset facilitates training multilingual models and benchmarking cross-lingual NLP.

Conclusions:

• The dataset meets practical demands for hate speech detection in English and Malay.
• It enables the development of effective multilingual hate speech detection systems.
• This resource supports educational NLP initiatives for English and Malay-speaking communities.