Intelligent Test Paper Generation Based On Genetic Algorithm Improved By Multi-Stategy

Recent years, using computer technology to reform the traditional examination manner, is a hot field of research and application of computer technology. An online examination system began to appear, with only by just converting the artificial arrangement of the paper into electronic version of the paper. And today there have been a lot of intelligent test systems that assemble automatic paper generation, paper scoring and test analysis. The twenty years of development, are based on issues of intelligent paper generation. Paper generation is a multi-constrained multi-objective optimization problem, the traditional algorithms based on mathematical algorithms have disadvantages of low speed of convergence, low success rate and paper quality, the papers produced often require manual adjustment. So the main purpose of this paper is to design a good algorithm to improve the efficiency and quality of test paper.This paper focuses on the application of genetic algorithm (GA) in multi-constraint optimization problems, and proposes an intelligent test algorithms based on GA. Practicing on the course "Computer Application", this paper first analyses the constraints of test paper in detail such as knowledge points, difficulty factor, discrimination degree etc., and then establishes a mathematical model of test paper generation, at last designs a kind of improved GA. The improved GA uses segmented matrix coding, the matrix maps the paper directly, and the each vector maps an question of the paper. The matrix vectors are divided into several segments, and the whole matrix mapped directly into the chromosome, can contain more information of the question, to facilitate calculated efficiency. Since the initial population contains the individual to meet the knowledge point, type of question, score and other constraints, reducing the number of iterations. Crossover and mutation operations are carried out in within segment, to make sure the right amounts match the right type of questions. The matrix with information about the questions can make the crossover and mutation are more selective, to avoid the problem of slow convergence brought by randomness. Finally, the algorithm is implemented and there is an experimental analysis, the results show that with the new method, success rate of paper generation and convergence speed have been significantly improved. It can also reduce the premature. The paper generated by the algorithm can better meet the requirements of the user, as long as the quantities of the questions are mature and reasonable.