Esker, a worldwide leader in AI-driven document process automation solutions and pioneer in cloud computing, today announced that it is presenting the results of its research on data extraction at the 2019 International Conference on Document Analysis and Recognition (ICDAR), the world’s top AI conference, in Sydney Australia. Also known as the “World Cup” in the field of document analysis and recognition, ICDAR is the biggest international gathering for researchers, scientists and practitioners in the document analysis community.
Clément Sage, a machine-learning PhD student and engineer at Esker, will present his paper titled “Recurrent Neural Network Approach for Table Field Extraction in Business Documents” in the context of his PhD research at Université Claude Bernard Lyon 1 in France.
Efficiently extracting information from incoming business documents, such as orders and invoices, is crucial for companies that face massive daily document flows. These documents contain valuable information that companies need to retrieve for integration in their ERP system and structured archiving. However, automating data extraction is extremely challenging, especially when analysing table content to identify ordered or invoiced items, as the documents have complex and ambiguous structures.
To address this problem, Esker developed an end-to-end method for table fields’ extraction by bypassing
physical structure recognition. Esker proposes a generic approach based on deep learning, which enables order item identification on layouts not necessarily seen during training. Sage’s first-time recognition approach is of particular interest to researchers, as it is generic enough to easily be adapted for other document types and relies on little domain specific textual preprocessing. This universal and nonproprietary AI algorithm is an important discovery for the document community.
Esker successfully evaluated the effectiveness of this approach on real-world purchase orders to retrieve product numbers, quantities and unit prices from ordered items. Following positive results, Esker is now using this technology on its cloud-based platform. The company’s customers are already benefitting from improved data recognition on new documents and better automation rates.
