ghananlpcommunity/multilingual-emotion-ghana
收藏资源简介:
一个多语言情感分类数据集,覆盖加纳的四种语言(英语、Twi、Ewe、Dagbani),旨在支持低资源非洲语言环境下的小型语言模型(如SmolLM2)的情感检测微调任务。数据集格式为指令调优(input → structured output),分为训练集、验证集和测试集(80/10/10比例)。每行数据包含input(指令提示、标签列表、语言标签和源文本)、output(结构化目标)、language(语言标识)和label(原始标签字符串)四个列。所有非英语样本都是相应英语样本的翻译,四个语言的子集在行数和标签分布上是平衡的。数据集的设计目的是用于微调小型语言模型,特别是在多语言非洲情感分类方面。
A multilingual emotion classification dataset covering four languages spoken in Ghana (English, Twi, Ewe, Dagbani), built to support fine-tuning of small language models (e.g. SmolLM2) for emotion detection in low-resource African language contexts. The dataset is formatted as instruction-tuned (input → structured output) and split into train, validation, and test sets (80/10/10 ratio). Each row contains four columns: input (instruction prompt with label list, language tag, and source text), output (structured target), language (language identifier), and label (raw label string). All non-English samples are translations of corresponding English samples, with balanced row counts and label distributions across all four language subsets. The dataset is designed for fine-tuning small language models, particularly for multilingual African emotion classification.




