Dipartimento di Ingegneria informatica, automatica e gestionale

Automated assessment of homework assignments is a challenging topic in programming courses. For some years we have been using our Q2A-II system, that supports 1) automated grading of homework programs, and 2) formative peer assessment, performed by students on the algorithm descriptions they submit with the homework. Here we present some of the data we collected during 4 years of activity with Q2A-II, in the framework of a university course on Basics of Computer Programming. Each year, 4 homework were administered to 300–500 learners, with about 8.600 submissions (each made of program + algorithmic description), overall, and about 23.300 peer evaluations. On such data we propose several observations, aiming to rate the effectiveness of the initiative, in view of a more in depth analysis. On the algorithm descriptions we performed a basic textual categorization, using BERTopic-based text embedding topic extraction. The classification aim is exclusively to assess whether a text can or cannot be considered as an algorithm description: in two of the research questions we try to validate the classification and to see how different can be the behavior of the authors of such descriptions during the peer assessment activity.