Research Of Auto-generating Test Paper Based On Reinforcement Learning

Test paper generating system is one of the important approach to realize the standardization, justice and rationalization in the examination. Different from just freeing teachers from the trouble of manually generating test paper, now focus has been changed to the research of generating algorithm and how to improve the quality of the test item and the efficiency of test paper generating system, with even broad application of artificial intelligent technology and computer technology. Based on analysis of plenty of relative articles in overseas or domestic, this article designed a common generating examination paper system on reinforcement learning, gave the new method to realize automatic generating examination paper. During the research, we realized some of the main algorithm, which is supported by the experiment result.At the beginning of this paper, I compared the efficiency of a group of test paper generating algorithems which included genetic algorithm and multi-goals evaluation function that are widly studied. According to the former analysis I disscuss a directive mapping approach method. The target of the directive mapping approach is to find a list of difficulty ranges, each of which relate to a uniquit test item in the test paper. The choice of the difficulty ranges should be random and its statistics should obey the nomarl distribution. The nomarl distribution is granteed by computation and analysis of requirment.The test system is composed of five parts: the quit part, the auto-generating test paper part, the test evaluation part, the score management part, the manual-generating test paper part. The auto-generating test paper part combines the reinforcement learning and auto-generating problem and use the feedback of reinforcement signal to train the system. Along with the accumulation of auto-generating experiment the efficiency of the system should be better and better.