A Multi-population Adaptive Genetic Algorithm for Test Paper Generation

Recently, online examination has well developped. As an essential task of online examination, test paper generation plays a critical role. Genetic algorithm has been used in test paper generation systems. However, traditional genetic algorithm is easy to be stuck at local optimal solutions and can not adapt well to the student's mastery of knowledge concepts. In this paper, we propose a novel Multi-population Adaptive Genetic Algorithm (MAGA) for test paper generation. Our algorithm introduces the Longest Common Subsequence(LCS) and Hamming distance to judge genetic similarity and comprehensively considers the influence of evolutionary generation, fitness and genetic similarity on parameters of the proposed algorithm. MAGA constructs the initial population which is then fed into the crossover and mutation operators to be optimized and produce the optimal solution. We conducted experiments for the sequential test paper generation on the datasets. Experimental results demonstrate that our algorithm achieves significant improvement over traditional genetic algorithms with faster convergence speed and higher fitness.