Constraint-based Problem Generation for a Self-Assessment System Nguyen-Thinh Le, Wolfgang Menzel Department of Informatics, University of Hamburg, Germany Abstract: In this paper, we describe a component of a self-assessment system which is capable to generate different exercise instances for different users. The system is used to give prospective students of Informatics a better feeling what kind of thinking skills this discipline requires. The system offers diversified problem tasks and by observing the performance of a student over a series of different problem instances we are able to assess the development of his/her problem solving abilities. We apply the constraint-based approach both for the problem generation and skill assessment. A first version of the system has been initially tested by thirteen high school students from Hamburg. Keywords: Self-assessment, constraint-based, problem generation Introduction Increasing demand of using information technology in industry, economy and administration results in brilliant career chances for Computer Scientists. Many prospective students ask themselves whether they should decide for a study with a major in Informatics. However, not many of them have a clear idea, which kind of skills are required for this discipline. No doubt, many people associate this kind of study with the ability of using computers, programming or creating spreadsheets. We developed a component INCOM-Self for a web- based system which helps prospective students to assess themselves whether their qualifications meet the requirement for a study of Informatics, namely working with formal descriptions, dealing with possibly conflicting requirements, as well as comparing alternative solutions with respect to their degree of goodness. Current surveyed self-assessment systems provide problem tasks for which usually a unique correct answer is expected and a solution has to be chosen from a list of alternatives. On the contrary, our self-assessment component poses questions which correspond more closely to real-life problems, namely problems with many possible solutions, or even no perfect solution at all. Then, the task of the student is to find an optimal solution for the given problem. We apply a constraint-based approach, both to generate problems and to evaluate solutions submitted by students. A constraint represents a requirement of a problem task. Each constraint divides the solution space into two parts: one part of correct solutions and another part of incorrect ones. All constraints taken together, describe the boundary for the space of correct solutions. Problem tasks can be generated based on a set of specified constraints at an almost arbitrary number and the student can be encouraged to practice the same type of exercise over and over again. Therefore, the system not only allows us to evaluate the problem solving capabilities properly, but also a student’s perseverance and learning progress. In the next section, we survey several self-assessment systems for prospective students of Informatics in Germany and review some techniques for problem generation. In the second section, we describe our self-assessment component and the constraint-based