HaEmoC-MLTC: Hausa Emotion Corpus for Multi-Label Text Classification from Twitter

The dataset comprises 12,761 Hausa tweets, each annotated with 11 distinct emotions: anger, sadness, disgust, fear, surprise, joy, trust, optimism, pessimism, anticipation, and neutral. The number of instances is relatively balanced across the emotion categories in the HaEmoC-MLTC dataset. It is specifically designed for multi-label classification tasks.

The tweets were collected via Twitter’s API, focusing on culturally significant events to capture a broad range of emotional responses. Native Hausa speakers manually labeled the tweets, ensuring high-quality annotations that accurately reflect the complex emotional expressions common on social media. Only the tweet text and corresponding annotations are released, in compliance with Twitter’s Terms of Service, which restricts the distribution of sensitive user data.

This dataset is essential for advancing emotion classification in low-resource languages and for supporting the development of more robust models for multi-label text-based emotion classification.