A comparative study on two large-scale hierarchical text classification tasks' solutions