NLPforLRLsProjects/SAfriSenti-Corpus-Mini

A multilingual sentiment corpus for South African Under-Resourced languages # SAfriSenti-Corpus-Mini A multilingual sentiment corpus for South African Under-Resourced languages. This corpus includes the following languages: Code-switched Dataset, English, Sepedi, Setswana, and Sesotho. We are working on extending the corpus to more languages like isiZulu, isiNdebele, isiXhosa, Tshonga, TshiVenda and Afrikaans. We aim to release the dataset in December 2026. We also aim to collect tweets from nearby countries like Lesotho, Botswana, Swaziland and the Shona language for Zimbabwe # SAfriSenti-Corpus Licenses This dataset is currently for a PhD project of Mr Ronny Mabokela enrolled at Wits University, South Africa. It will be licensed and approved within the University. An ethical clearance has been obtained and approved for this research. # SAfriSenti-Corpus Focus This project focuses on developing sentiment resources for South African languages. The dataset will be available for the NLP community for low-resource languages. # Training Data We provide the training data for each language, including code-switches, separately. # Some Results with African LLMS ``` Accuracy: 0.7832796713667147 Classification Report: precision recall f1-score support 0 0.78 0.90 0.83 4583 1 0.59 0.46 0.52 668 2 0.83 0.69 0.75 3756 accuracy 0.78 9007 macro avg 0.73 0.69 0.70 9007 weighted avg 0.78 0.78 0.78 9007 Classification Report: precision recall f1-score support 0 0.68 0.73 0.71 682 1 0.80 0.82 0.81 3820 2 0.85 0.83 0.84 4505 accuracy 0.82 9007 macro avg 0.78 0.79 0.79 9007 weighted avg 0.82 0.82 0.82 9007 Classification Report: precision recall f1-score support 0 0.53 0.68 0.59 671 1 0.45 0.19 0.26 707 2 0.64 …