Parsoffensive: Persian Offensive Comments Dataset

Offensive language detection is essential for maintaining respectful and inclusive online environments. By implementing detection systems, platforms aim to protect users from harmful content, curb the spread of hate speech, and promote positive interactions within digital communities. Leveraging natural language processing (NLP) techniques, this area has become a key focus in computational linguistics research and development. To address this challenge, we created the ParsOffensive dataset, consisting of approximately 8,433 Persian comments labeled as Offensive or Neutral. These comments were sourced from Instagram and labeled by two linguists. Machine learning models were trained on this dataset, with the ParsBert model achieving 90% accuracy in detecting offensive speech in Persian.