Scheduling Software for Improving Teaching Assistant Schedule Satisfaction

Abstract

Teaching assistant scheduling at Chalmers University of Technology and the University of Gothenburg is often managed through manual and unstructured processes, such as spreadsheets and informal communication, leading to inefficiencies, scheduling conflicts, and unfair workload distribution. This bachelor’s thesis investigates how an Automated Teaching Assistant Allocation System can improve teaching assistant schedule satisfaction by combining algorithmic optimization with user defined constraints and preferences. The project focuses on both the technical problem of generating feasible schedules and the user-centered challenge of supporting teaching assistants in expressing availability and preferences in a clear and usable way. To address this problem, a prototype web-based scheduling system was designed, implemented, and evaluated. The system allows course responsibles to configure courses and sessions, while teaching assistants can specify hard constraints, such as unavailable times, and soft constraints, such as preferred session types and scheduling preferences. Several scheduling algorithms were implemented and compared, including Constraint Programming models using Choco Solver with Large Neighbor hood Search, and a hybrid approach combining Google Operations Research-Tools Constraint Programming-Satisfiability with a heuristic greedy algorithm. The algorithms aimed to satisfy all hard constraints while minimizing penalties associated with violated soft constraints. A survey with 40 teaching assistants, an interview with a course responsible, and usability interviews with four teaching assistants were conducted to inform both system design and evaluation. The results indicate that current scheduling approaches are perceived as time-consuming, inconsistent, and lacking support for fairness and preference satisfaction. User evaluations of the prototype showed that participants found the constraint input process relativelt intuitive and expressed a preference for the system over current scheduling methods. Benchmark testing demonstrated that Large Neighborhood Search-based approaches improved scheduling quality compared to a naive implementation, particularly for larger scheduling problems. The results suggest that combining constraint based optimization with a user-centered interface can support both feasible scheduling and improved teaching assistant satisfaction.