Application of Swarm-intelligent Methods to Optimize Error-tolerant Graph Matching for Automatic E-Assessment
Tobias Reischmann, Breno Augusto de Melo Menezes · 2019
Automatic e-assessment poses the challenge to design algorithms, which are capable of identifying patterns within a student's submission to distinguish the correct and incorrect aspects of their solution. Error-tolerant graph matching techniques using graph edit distance metrics are one of the applicable methods. There, a reasonable identification rate is only achievable if we use a proper cost function for the edit operations. In this paper, we optimize the matching quality of an e-assessment system, which identifies design patterns within a UML class diagram as part of the automatic analysis of the student's submission. For this purpose, we apply particle swarm optimization and fish school search to optimize the parameters of the cost function by comparing the matching results to the assessments of students and teachers. We further show the complexity of the resulting fitness function and justify the application of swarm intelligent heuristics for such application scenarios.