Vortrag: Entity Recognition and Accounting Type Prediction


Simplifying Accounting on the Basis of Machine Learning

sevDesk is an online office and accounting software that allows users to create invoices, track expenses, document contacts and manage products. We try to simplify repetitive chores for entrepreneurs and small businesses for them to be able to focus on their actual work.

A power user of sevDesk uploads more than 150 credit and debit receipts every month. While all relevant information is already printed on the receipt, it still has to find its way into our database to be processed later when the user wants to create balance sheets and VAT return. Instead of asking the user to manually enter the data, we exploit the relatively rigid  structure of invoices to automatically filter out relevant information like amount, invoice date or company name. However, it does not suffice to simply use optical character recognition along with regular expressions and XY-coordinates to get satisfying results. Instead, we train machine learning models like conditional random fields and recurrent neural networks to obtain a named entity recognition model that is able to generalize to unseen data. In a next step, we use the extracted data to predict the accounting type of each receipt.

We present the end-to- end process from data collection and annotation to model deployment in this talk.

David Bläsi – SEVENIT GmbH

David Bläsi (SEVENIT GmbH)David Bläsi hat an der Albert-Ludwigs-Universität in Freiburg Mathematik studiert. Seine Studienschwerpunkte lagen in der Wahrscheinlichkeitstheorie, seine Masterarbeit schrieb er am Institut für medizinische Biometrie und Statistik in Freiburg. Seit 2017 ist er Teil des Machine Learning Teams bei der Sevenit GmbH in Offenburg und beschäftigt sich dort mit Entity Recognition und Document Classification.

 

 

 

Hasham Munir – SEVENIT GmbH

Hasham Munir (SEVENIT GmbH)

Hasham Munir is working as a data science engineer in machine learning team of SEVENIT GmbH since 2016.He is responsible for handling the ‘Accounting type Prediction’ using machine learning techniques and creation of APIs and Automation processes. He has completed his MSc. In communication and media Engineering from Offenburg University of Applied Sciences. He also has developed fraud detection system for Belgium based company. He has been awarded twice from Daad and Lionsclub during his educational tenure.