Кросс-документальное семантическое улучшение

AbstractThe introduction of multimodal information and pretraining technique significantly improves entity recognition from visually-rich documents. However, most of the existing methods pay unnecessary attention to irrelevant regions of the current …...