Investigation On The Application Of Test Paper Auto-generation System Based On Improved Genetic Algorithm

Test paper auto-generation is to generate a test paper with specific total point, total time, point's distributions of difficulty, distinction, knowledge, type, cognition and so on according to a certain requirements by automatically selecting questions from bank of test questions. Test paper auto-generation is one of the most important components of Computer Managed Instruction (CMI). Several algorithms have been applied in test paper auto-generation, such as priority strategy, randomization, and trace and trial strategy, genetic algorithm and so on. When encountering with big solution space, multimodal problems, these algorithms are usually inclined to run into local extremum or difficult to solve. Since generated test paper needs to fulfill variable demands with great randomness and rationality, a more effective algorithm is in great demand. The thesis is to develop an improved genetic algorithm, and apply the algorithm in the test paper auto-generation problem. The main works in this paper are as follows:Based on the fundamental principles of test paper generation, analyses of evaluating indicators and their functions, relationship were done in detail; a test paper generation mode was set up with the distributions of the evaluating indicators. Finally, a mathematic modal to generate test paper was established grounded upon the preference of test paper quality defined by the generation mode.A new heuristic adaptive genetic algorithm based on niches was developed and investigated. An adaptive constant C_min is introduced according to the variety of population fitness distribution to deal with the prematurity and the low convergence speed of genetic algorithm. By adjusting C_min, the selection probability of every population was optimized. Several classical test functions for global optimization were adopted to validate the effectiveness. The test results show that ANGA can improve global optimization ability and the convergence speed, and the algorithm is quite robust.The procedure to solve the problem of test paper auto-generation with the original ANGA is presented in detail, and the test paper auto-generation of the present "National computer rank examination (3): information management technical" was taken as an example. The tests results indicate that: ANGA can be successfully applied in the test paper auto-generation system; the speed is quite fast, the success rate is great, the sensitivity toinitial value is dull.After the requirements analysis done in depth, Unified Modeling Language was adopted to analyze and design the online test system, an interface friendly online test system was accomplished.