Research On The Problem Of Auto-Generation Test Paper Based On Hybrid Genetic Algorithms

Recently, with the development of computer-Based Education, Web-based Education Resource Management System as an important component of Computer-assisted Testing gets more and more attention. In order to comply with national education trends for the modernization, network education resource management system as an important component of modern education system has been considerable development in the country. In the examination system, test paper auto-generation is an important factor in evaluating the system and decides whether the system can effectively test students' truth levels. So the study of test paper generation algorithm is a significant topic of computer based education. Auto-generation test paper is pre-set in accordance with the constraints and automatically generated by the computer, which meet the following constraints: knowledge structure, difficulty, degree of differentiation, cognitive level, time, score, etc. Auto-generation test paper implementation process involves algorithms and data structures, mathematical modeling, optimization control, artificial intelligence and many other fields of technical problems. Auto-generation test paper addressed the key issue is bound to satisfy the optimum conditions Existing test paper generation algorithms have some defects, such as low success ratio, costing long time and poor quality of test paper. Aimed at these defects, the intelligent test paper auto-generation algorithm based on hybrid genetic algorithm is researched in this paper.The test is the important part of the whole process in the existing educational system, , which is an important factor to evaluate the teaching effectiveness. The perfection of the examination item bank system plays a very important role in enhancing the standard of teaching, teaching effectiveness,and the scientific, efficient and rational auto-generation test paper system also plays a particularly crucial role in examination system. With the advance of computer-aided education, the traditional examination proposition not only spends considerable time and effort, but also can't meet the objective needs of the current Web-based Education. In addition, the manual test paper often make mistakes because the teachers' subjective ideas about question properties, so it is hardly test the students' truth levels and evaluate the quality of the process. Taking the above the two factors can be seen that a good test paper system for the whole process of teaching plays an important role. The purpose of this article is to solve defects of the existing test paper generation algorithms such as low success ratio, costing long time and poor quality of test paper.Firstly, the basic theories and principles of paper generation are described, the advantages and disadvantages of classical test theory and item response theory are evaluated. The constraint conditions are summarized. Based on them, a mathematical model of test paper generation is established and the objective functions are proposed. And then this article briefly describes the need to use the relevant test paper algorithms: genetic algorithms, local search algorithm, tabu search algorithm, multi objective genetic algorithm and genetic algorithm optimization and other issues. Secondly, aimed at genetic algorithm's shortage, the reason why precocious convergence is easy to occur, usual solutions to precocious convergence and measurements of population diversity are introduced. The algorithm uses the Sub-integer-coded based on the types of questions. Three genetic operators adopt the following strategy, the ( ? ? ?)selection strategy of combinatorial optimization evolutionary algorithm is adopted by selection operator, and the whole population is divided into sub-populations by types of questions. In each sub-population ? parents generate ? offsprings by genetic operators. Then all individuals are gathered and sorted, the fitness sharing is performed and better individuals are selected to be the next generation. A class of parallelism evolution technique for niches is implemented by crossover operator according to the types of questions, which can ensure the total numbers of selected questions in each type are not changed after this operation. It was proved theoretically and analytically this kind of niche technique can provide strong selected pressure and also maintain the diversity of individuals in populations, and it can remarkably improve the reliability of global convergence and converging velocity. Firstly, the median variation should be calculated. Secondly, the tabu search algorithm will be called in order to search for new individuals to replace the old individuals. In the early stages of search, a random search strategy of larger probability should be adopted in order to improve the global search performance of the algorithm. In the later stages of search, a greedy search strategy should be adopted in order to speed up the convergence of local search process. The validity of the improved algorithm is proved by simulation tests. The tests results show that the improved algorithm can improve global optimization ability and the convergence speed.The improved algorithm is used to solve test paper generation problem. The process of software system is described by the UML, which includes needs analysis, overall design, detailed design, coding, implementation, testing, and several other aspects, which based on Rational Unified Process theory. The simulation test indicates the intelligent test paper auto-generation algorithm based on improved hybrid genetic algorithm can satisfy the needs for actual examinations, the speed is faster and the quality of test paper is better.