GFTE: Graph-Based Financial Table Extraction

Tabular data is a crucial form of information expression, which can organize in standard structure for easy retrieval and comparison. However, financial industry many other fields, tables are often disclosed unstructured digital files, e.g. Portable Document Format (PDF) images, difficult to be extracted directly. In this paper, facilitate deep learning based table extraction from we publish Chinese dataset named FinTab, contains more than 1,600 diverse kinds their corresponding representation JSON. addition, propose novel graph-based convolutional neural network model GFTE as baseline future integrates image feature, position feature textual together precise edge prediction reaches overall good results https://github.com/Irene323/GFTE.