Text this: Question-answering enhancement method for large educational models based on re-ranking and post-retrieval reflection