2023: Generalizing Machine-Learning Based Assessments

Master's theses

Paul Schwind



Learning Management Systems like Artemis have become pivotal in administering programming courses as educational platforms evolve to accommodate modern learning paradigms. However, tutors still face the time-consuming task of providing detailed feedback. While the recently proposed Athena system has begun to semi-automate this process for text exercises, a gap remains for programming exercises. The research in this thesis aims to fill that void by adapting Athena to handle both text and programming exercises in a unified manner. We present a modular design that streamlines the addition and interchangeability of feedback-generating components. This new architecture includes CoFee, the current text-exercise feedback module from Athena, and a machine-learning module specialized for programming exercises. We integrate this new Athena system into Artemis. An analysis of Athena’s existing framework reveals areas for improvement. We design and implement a modular structure to support multiple types of exercises and allow for the addition of new feedback modules. A brief evaluation follows to assess the quality of the system’s automated feedback. By expanding Athena’s capabilities, this thesis aims to enrich the feedback loop in programming education, thereby offering advantages to both tutors and students.

Head of Research Group

Prof. Dr. Stephan Krusche

Administrative office

Sophia Adelmeier

Office: 01.07.052
Phone: +49 (89) 289-18206
Fax: +49 (89) 289-18207 Boltzmannstraße 3
85748 Garching b. München