We are aiming to extract every part of the 5W1H, but not necessarily every slot can be filled, since for instance the answer to how something happened might not be given in the text.
Here We divide our approach into six sub-tasks and group them in three steps: (1) Title classification and topic sentences extraction for key event identification; (2) Semantic role labeling and 5W1H elements identification for event semantic elements extraction; (3) Collect extracted event facts automatically according to nature of existence.