Submissions will be evaluated based on accuracy, which measures the proportion of correctly predicted emotion labels out of all test samples.