Hybrid Genetic Adaptive Test Paper Generation Algorithm with Deep Learning for Online Testing Systems

Abstract

Socioeconomic development has brought about the expansion of education scale. Due to the tasks of printing test papers, proctoring, and grading, teachers' workload is increasing, and these processes are all manually completed, which is very prone to er rors. Consequently , this paper proposes an online test system that integrates deep learning and an automatic test composition algorithm. The objective of this integration is to automatically generate test papers that align with teachingrequirements throug h intelligent mean s. This approach aims to minimize manual intervention, thereby reducing error rates and enhancing test efficiency. Firstly, the paper studies the automatic paper composition algorithm based on deep learning and optimizes the paper composi tion efficiency by intelligently reducing the search space, setting constraints, and using real number segmentation coding. Then, an online test system is designed to automatically record answers and score them. Finally, the results of the online test syst em combining deep learning and automatic test composition algorithm are analyzed. The results showed that when the evolutionary algebra reached about 400, the population maximum fitness value of the studied algorithm was stable at 18. In 20 independent exp eriments, the algo rithm showed excellent convergence performance. The convergence algebraiccurve of the first experiment tended to be stable in about 180 generations, and the shortest runningtime of the algorithm was only 0.5 seconds. The average accurac y of the research algorithm was 92%in the difficult test task, and 93% in the test task, which fully verified the efficiency and stability of thealgorithm. With the increase in population size, the mean fitness of the online test system proposed bythe s tudy also increase d. When the individual population size was 300, the mean fitness of the onlinetest system proposed by the study was 17.173, the mean fitness of the exam treasure was 17.162, andthe mean fitness of the exam star was 17.158. The mean fitn ess of learning Xi aoyi was 17.153. Thesuperior optimization ability and efficient convergence performance of the online test system canprovide strong support for the rational allocation of educational resources and the realization ofpersonalized teachin g