DGCN Based Solution for Entity Linking on Visual Rich Document

Abstract

Various works on entity extraction on visual rich document (VRD) have been done. However, few methods have been explored to handle entity linking problem. The difficulties come from the number of possible linking edges among entities is of square times complexity. Our approach introduces directed graph based convolutional network (DGCN) to predict relations between entities, which out performs existing methods on the FUNSD entity linking task.

Tasks

Reproductions