Research On Efficient Auto-generating Examination Paper Algorithm And Personalized Exam

Examination,the conventional learning detecting strategy,has been proved to be a practical approach.Nonetheless,conventional assemble exam paper approaches cannot fulfill the exponentially increased demand for exams,due to its low efficiency and limited ability to personalization,within the circumstances of the rapid development of Internet education.This drawback becomes one of the bottlenecks obstructing the development of Internet education.As such,to achieve high-efficiency exam paper assemble and provide personalized detection solutions for users is an urgent task.In order to exhaustively examine the knowledge level of a user and generate personalized test papers for different users efficiently,this thesis did several works as follows:1.We first analyzed the current research trend and shortcomings of auto-generating examination papers domain.Then for a set of problems in conventional algorithms,i.e.,high assembly time,low efficiency and inability to satisfy the multi-conditions,we propose a multi-pack container auto-generating examination paper which based on knapsack problem and greedy algorithms.2.According to current recommendation strategy,we introduce a recommender approach based on auto-generating examination paper.Our recommendation models are optimized by gradient descent,and trained to generate personalized exam paper for users in different knowledge level.3.Trend concentration phenomenon is a general problem in recommender system filed.To tackle this issue,we utilize a differential mechanism adding Laplacian noise to our recommendations.This thesis experiments aforementioned algorithms respectively.In the experiment of auto-generating examination paper,we compared our method with conventional algorithms.The results of 62.05% improvement of efficiency reveal the time consuming of our algorithm is far less than traditional algorithms,within complicated assemble condition.On the other hand,in recommendation experiment,we propose a set of recommendation strategies on different recommender tasks to verify both recommendation and trend concentration problem.The results illustrate each strategy provide reasonable consequences.The prediction error drop obviously,which proves the feasibility,universality,and accuracy of the model.By comparing with the precision and recall,results of the recommendation experiments demonstrate the recommender model which added Laplacian noise prevent the occurrence of trend concentration phenomenon while guaranteeing the accuracy of recommendation.As such,compared with traditional assemble exam paper model,our model is more feasible and efficient.