【作者】 刘超；

【导师】 胡燕；

【作者基本信息】 武汉理工大学 ， 计算机科学与技术， 2017， 硕士

【摘要】 随着信息技术的发展,在线考试系统应用越来越广泛,智能组卷作为在线考试的核心,对其的要求越来越高。智能组卷算法的设计很大程度上决定了生成试卷的质量和效率。遗传算法不受搜索空间限制性条件约束,简单、鲁棒性强,并能进行全局寻优,被广泛应用到组卷中。但是它同样存在着早熟以及局部搜索能力差等缺点。本文基于遗传算法对智能组卷方法进行研究,主要完成了以下工作。首先,以经典测试理论为基础,对试题以及试卷的指标属性进行了分析,并简要介绍了部分指标间的相互关系,完成组卷评价指标的设计工作。其次,构建智能组卷的数学模型,计算用户期望与生成试卷各指标之间的差距,进而确定组卷的目标函数。之后,引入BP神经网络,对智能组卷目标函数的权重系数进行训练优化,进而更科学地确定目标函数。然后,对遗传算法进行改进。在选择操作中采用了一种基于轮盘赌的多轮轮盘赌选择法优选个体。为了保证算法的收敛性,使用最优个体保留机制。路径选择,即以不同的概率选择进行交叉操作或者变异操作,加快算法收敛速度。自适应调整交叉概率以及变异概率。将变异算子分成自适应算子和速成算子两部分。引入小生境技术,保持种群多样性。实验表明改进遗传算法能够较好的解决典型函数问题,特别是对多峰函数的处理。最后,将改进遗传算法应用到组卷系统中,并通过实验对组卷效果进行评价。实验结果表明,使用改进算法的组卷系统能够解决智能组卷问题,达到用户的预期。更多还原

【Abstract】 With rapid development of information technology,online examination system is more and more widely used.As the core of online exams,Intelligent Test Paper Generation(ITPG)demands better performances for efficiency and quality of automatically test paper generating,which however is decided by the design of its algorithm.Genetic algorithms are simple and robust algorithms with global optimization capability,which are not constrained by search space constraints.They are suitable for ITPG.But there are obvious deficiencies of Genetic algorithms,such as immature convergence and short of climbing ability.This thesis researches ITPG methods based on Genetic algorithms and mainly completes the following work.Firstly,the index attributes of questions and papers are analyzed according to the classical test theory,and the relations of certain indexes are briefly introduced,with an assessment criterion.Secondly,the mathematical model of ITPG is proposed.The difference between user expectation and the index of the test paper is calculated according to ITPG’s mathematical model,and then the objective function of the test paper is constructed.We present an enhanced method based on BP neural network to optimize the weight coefficient of objective function in ITPG model.Thirdly,a modified genetic algorithm is proposed.We adopt a multi-wheel roulette selection method based on the roulette method as the selection operation to optimize the individual.The optimal individual is reserved to ensure the convergence of the algorithm.Routing selection selects the crossover operation or the mutation operation with different probabilities to speed up the convergence rate of the algorithm.Crossover rate and mutation rate are adjusted adaptively.A reconstruction mutation operator is designed with two parts: crash operator and adaptive operator.Niche technology is imported to maintain population diversity.Experiments demonstrate that the improved genetic algorithm can solve the problem of typical function,especially the processing of multimodal function.Finally,the improved genetic algorithm is applied to ITPG system,and experiments are designed to evaluate the enhanced ITPG system.The experimental results show that the proposed ITPG system is effective and can meet the user’s expectations.更多还原