annotation-consistency

This study evaluates LLMs as data quality annotators on e-commerce tasks, finding they outperform baselines when background knowledge is required but offer limited advantages for tasks with strong lexical signals, while demonstrating high consistency across runs.