Mabast dataset for Sorani Kurdish dialect semantic text classification

Mabast is a labeled Sorani Kurdish Dialect semantic text classification dataset developed to support Kurdish Natural Language Processing (NLP), particularly text classification and harmful-content detection in a low-resource language. The dataset was prepared with the contribution of volunteer participants and refined through a multi-stage curation process. Entries were reviewed to ensure that they were readable, written in Sorani Kurdish Dialect, relevant to one of the five semantic categories, and suitable for annotation. Non-Sorani, unclear, duplicated, empty, and irrelevant entries were excluded. The text samples were cleaned, normalized, checked for spelling and consistency, deduplicated, and verified for missing values before final release. The reviewed samples were annotated according to five categories: Normal, Romantic, Advice, Threat, and Hate Speech. The released Comma-Separated Values (CSV) dataset contains 19,922 labeled text samples with two columns: text and numerical label. A baseline validation experiment using Term Frequency-Inverse Document Frequency (TF-IDF) features and a balanced linear Support Vector Machine (SVM) classifier is provided as a reproducible evaluation setup. The dataset can be reused for Sorani Kurdish Dialect semantic text classification, harmful-content detection, and the evaluation of machine learning and deep learning models for low-resource Kurdish NLP.