Project DeepWrite • University of Passau

DeepWrite focuses on interdisciplinary research on and the development of AI-based assistance systems, that aim to support students with strengthening their writing and argumentative skills in the fields of economics and law. It aims at rating texts written by the students in terms of structure and argumentative quality through an automated process.

The following institutions are involved in the project: Lehrstuhl für Data Science (Prof. Dr. Michael Granitzer), Forschungsgruppe CAROLL (Prof. Dr. Jelena Mitrović), Lehrstuhl für Volkswirtschaftslehre mit Schwerpunkt Wirtschaftstheorie (Prof. Dr. Johann Graf Lambsdorff), Prof. Dr. Marcus Giamattei, Lehrprofessur für Öffentliches Recht am Institut für Rechtsdidaktik (Prof. Dr. Urs Kramer), Zentrum für Informationstechnologie und Medienmanagement and the Didaktisches Innovationslabor.

Technical Solution

1: Struktur- und Sprachanalyse; 2: Antwortbewertung: Form, Stil, Semantik; 3: Fachbezogener Wissensgraph (Rechtsontologie, Wirtschaftswissenschaften); 4: Erstellung aus offenen Daten, Erweiterung durch Fachexperten; 4: Rückmeldung an das System durch Lehrende; 5: Rückmeldung an das System durch Studierende — Darstellung: Prof. Dr. Michael Granitzer, Lehrstuhl für Data Science

(1) By using transformer networks in combination with attention models (a special deep learning method) DeepWrite is supposed to extract discourse graphs from students ‘answers and sample solutions. (2) Through these discourse graphs the structure of students ‘answers and sample solutions can be analysed and connected to the knowledge graph. The discourse graphs act as a normalisation of the language and is meant to connect the arguments with the domain specific concepts in the knowledge base. (3) The semantic knowledge graphs stem from public resources (i.e. Wikipedia) and analysed materials (i.e. legal texts). (4) They are then improved by lecturers. (5) Feedback from lecturers as well as from students can help to improve the system through an active learning process.

Team

Mitarbeiterinnen und Mitarbeiter des Projekts: Paul Brudny, Walter Doyle, Prof. Dr. Urs Kramer, Alberto Mader, Carina Trauner, Silvia Gürtner und Sarah Großkopf — Bild: © Melanie Bauhuber

Introduction (11:30).

Funder: “Digitale Hochschulbildung” by the German ministry of education and research; federal budget, single plan 30, chapter 03, Title 68518, fiscal year 2021

Project name: KI-gestützer Erwerb von Schreib- und Argumentationskompetenzen in den Disziplinen Jura und Wirtschaft – DeepWrite

Subject areas according to the DFG-classification: law (113), economics (112), informatics (409-06)