F1 score is one of the metrics used to evaluate reading comprehension systems. It takes the average word/token overlap between predicted and gold answers. Treat the prediction and gold as a bag of tokens, and compute F1 for each question, then return the average F1 over all questions.