Groupwork is believed to have positive effects on learning. This study addresses a practical use-case where students solve assignments in groups yet demonstrate their learning outcomes through a portfolio of individual reports. When grading such coursework, it is beneficial to collectively examine the individual report written by all the group members. However, students often make mistakes when declaring their group composition, while others forget or refuse to disclose their team. This can be particularly challenging with anonymized portfolios. This paper presents a simple, accurate, and computationally efficient unsupervised algorithm for detecting group structures based on the contents of the reports. Tests on 2151 reports by approximately 500 students showed that a large portion of groups could be identified successfully (33%–88%), with a mean similarity of 63%–94%, mean precision of 83%–97%, and mean recall of 70%–97%. The approach thus holds potential as a pragmatic tool for administering coursework assessments. Details on how to adjust the approach for three additional use cases are presented, namely the matching of final and former reports by the same students, identifying potential plagiarism cases for manual screening, and identifying incorrectly submitted reports.