Make the criteria visible
Questionnaire items describe skills, experience, learning goals, and working styles, with each criterion marked for Similarity or Diversity.
Work
Algorithmic Group Formation
Transparent group formation workflow for a studio of 84 architecture students, combining questionnaire-based Similarity and Diversity scoring, Python assignment, instructor review, and an optional AI-assisted questionnaire extension.
Stage
How can an algorithm support group formation?
The research develops a transparent workflow for forming groups in an 84-student architecture studio through questionnaire-based Similarity and Diversity scoring, Python assignment, and instructor review.
Score every possible pair
Python evaluates all 3,486 unique pairs, combines the scoring matrices with adjustable weights, and prepares a draft allocation under group-size constraints.
Keep the instructor in the loop
The workflow exposes its criteria, weights, and diagnostics so instructors can inspect and revise the result rather than accepting a black-box decision.
Develop it through two academic years
The method grew through teaching in 2024-2025 and 2025-2026 and became a conference paper scheduled for publication in 2026.
Backstage
The hard part was making the rules useful.
I had to translate a teaching question into explicit criteria without pretending that the algorithm could replace instructor judgment or student agency.
I separated similarity from diversity
Some criteria benefit from common ground while others benefit from complementary strengths. Keeping the matrices separate made those choices easier to inspect and adjust.
I treated the first allocation as a draft
The script proposes groups and reports its reasoning. The instructor still checks the result, changes weights when necessary, and reruns the assignment.
I learned from the classroom deployment
All 84 students completed the final questionnaire. No transfer request was recorded apart from one administratively handled case, which kept the evaluation grounded in the actual studio.
I added AI only where it helped
The later OpenAI API extension suggests editable questionnaire items from a course brief. It does not assign students, and every suggestion remains under instructor review.