Online Peer-Assessment Datasets

TL;DR

This paper introduces datasets from semester-long peer-assessment experiments among university students, designed for use in NLP and ML research, focusing on student performance and answer evaluation.

Contribution

It provides detailed, structured datasets from real peer-assessment activities, enabling diverse experiments without expert ratings.

Findings

Datasets include student scores and submitted questions.

Designed for NLP and ML applications.

Supports performance prediction and text similarity tasks.

Abstract

Peer-assessment experiments were conducted among first and second year students at the University of Trento. The experiments spanned an entire semester and were conducted in five computer science courses between 2013 and 2016. Peer-assessment tasks included question and answer submission as well as answer evaluation tasks. The peer-assessment datasets are complimented by the final scores of participating students for each course. Teachers were involved in filtering out questions submitted by students on a weekly basis. Selected questions were then used in subsequent peer-assessment tasks. However, expert ratings are not included in the dataset. A major reason for this decision was that peer-assessment tasks were designed with minimal teacher supervision in mind. Arguments in favour of this approach are presented. The datasets are designed in a manner that would allow their utilization in a variety of experiments. They are reported as parsable data structures that, with intermediate processing, can be moulded into NLP or ML-ready datasets. Potential applications of interest include performance prediction and text similarity tasks.