llm security, datasets, efficiency, abusive language detection, temporal information extraction, social media processing, danish