Information Extraction from Invoices: A Graph Neural Network Approach for Datasets with High Layout Variety

Research output: Contributions to collected editions/worksArticle in conference proceedingsResearchpeer-review

Authors

Extracting information from invoices is a highly structured, recurrent task in auditing. Automating this task would yield efficiency improvements, while simultaneously improving audit quality. The challenge for this endeavor is to account for the text layout on invoices and the high variety of layouts across different issuers. Recent research has proposed graphs to structurally represent the layout on invoices and to apply graph convolutional networks to extract the information pieces of interest. However, the effectiveness of graph-based approaches has so far been shown only on datasets with a low variety of invoice layouts. In this paper, we introduce a graph-based approach to information extraction from invoices and apply it to a dataset of invoices from multiple vendors. We show that our proposed model extracts the specified key items from a highly diverse set of invoices with a macro F 1 score of 0.8753.

Original languageEnglish
Title of host publicationInnovation Through Information Systems - Volume II : A Collection of Latest Research on Technology Issues
EditorsFrederik Ahlemann, Reinhard Schütte, Stefan Stieglitz
Number of pages16
Place of PublicationCham
PublisherSpringer Science and Business Media Deutschland GmbH
Publication date2021
Pages5-20
ISBN (Print)978-3-030-86796-6
ISBN (Electronic)978-3-030-86797-3
DOIs
Publication statusPublished - 2021