how LayoutLM aggregating words as semantic entities

[kalelsun]

I have learned how to use LayoutLM for word-level classification. However, I am unsure how to aggregate words to form an entity. Can someone please provide guidance or resources on this matter

[yang0369]

one idea is to aggregate them by a rule-base algorithm,e.g. you can merge the B-question, I-question and e-question if they are close in position and by correct sequence. It should be quite easy and accurate doing so.