Abstract
In recent years, online hate speech has posed a growing threat to user safety, social harmony, and community cohesion especially on social media platforms. However, most existing hate speech datasets are monolingual and resource-rich, which leave Southeast Asia languages such as Malay underrepresented in natural language processing research. The aim of this dataset is to handle this gap by providing a balanced and quality-controlled resource that supports machine learning applications in multilingual settings. Binary classification is selected as a foundation task because it simplifies practical deployment in real-world and early-stage detection systems. It is beneficial in low-resource languages where detailed or multi-label annotations are always unavailable or inconsistent. This dataset presents 26,985 bilingual Malay-English social media texts curated from five public sources for binary hate speech detection. It combines human-annotated and filtered through controlled pseudo-labelling to retain only high-confidence, quality-controlled texts. The dataset is provided in UTF-8 encoded CSV format with 13,609 English and 13,376 Malay-language texts. Each entry includes the social media post, binary label (0 = non-hate, 1 = hate), language identifier (en or ms), and data source information. The dataset meets clear practical demands, including training multilingual transformer-based classifiers, benchmarking cross-lingual NLP models, and developing effective hate speech detection systems and educational NLP resources for English and Malay-speaking communities.