Key Methods And Applications Of Intelligent Learning Diagnosis For Intelligent Evaluation

Since the 21st century,the development of "Internet+Education" has become more and more mature,and users’ learning demands have gradually shifted toward intelligence.At the beginning of 2020,the impact of the epidemic has caused large-scale traffic to flow into the online,and more users will be exposed to smart learning.At the same time,with the support of theoretical support and technology drive in the field of education and information technology,smart learning has shown a strong growth trend,and the market scale and user scale have grown at a high speed,and a variety of intelligent evaluation modes represented by intelligent cognitive diagnosis have emerged.Intelligent learning diagnosis makes full use of cognitive diagnosis technology,and mines learners’ cognitive level by constructing cognitive diagnosis models,so as to carry out targeted learning interventions for learners.At the same time,Its intelligent learning method gets rid of the geographical restrictions of high-quality teaching resources,allowing learners from different regions to enjoy personalized and diversified educational resources according to their own learning conditions,which relieves educational inequity to a certain extent.While intelligent learning diagnosis breakes the traditional concept of "focus only on scores",its diagnosis results also provide a reliable decision-making basis for the adaptive learning engine,and strive to achieve teaching in accordance with their aptitude.It can be seen that the intelligent learning diagnosis technology is burgeoning,but it cannot be ignored that it still has the following problems:the item attribute pattern is not well quantified;the granularity of cognitive diagnosis model is relatively coarse and lacking of cognitive diagnosis model of combined subjective and objective items;manual labeling of missing item attribute pattern is difficult and subjective factors have a great influence;parametric modeling and marking methods are highly complex,in addition,only the influence of a single factor on learners is considered in the personalized test paper formation based on the analysis results,and the learner’s individual learning needs cannot be met.In response to the above problems,this research mainly starts with the construction of cognitive diagnosis models and the prediction and correction of item attribute pattern in two scenarios,and combines the personalized test paper generation strategy in practical applications to conduct the following research:(1)In the case that the item attribute pattern is known in the small and medium-scale evaluation scenarios,the research is aimed at the problems that the quantitative representation relationship of the item attribute pattern needs to be improved,and the intelligent learning diagnosis lacks cognitive analysis,and lacking of cognitive diagnosis model of combined subjective and objective items.(a)Constructing the tensors of test,knowledge,and ability,which gives traditional item attribute pattern with pedagogical interpretable information,and introduces education domain knowledge from the data layer for interpretability modeling.(b)Taking the Test-Knowledge-Ability tensor as the initial item attribute pattern,estimating the model parameters based on the expectation maximization algorithm,analyzing the learner’s potential cognition based on the maximum posterior probability method,and developing an interpretable multilevel cognitive diagnosis model for objective items.(c)Based on the Markov chain Monte Carlo algorithm,a multilevel cognitive diagnosis method for joint modeling of subjective and objective items is developed,so as to mine the learner’s knowledge and cognitive level,refine the granularity of the mining results,and provide reliable data support for the learners’ learning and analysis results with strong explanatory information.Finally,experiments prove that the average attribute match ratio of the multi-level cognitive diagnosis model reaches 91.6%,and the joint diagnosis model for subjective and objective item is better than the existing model and has a finer diagnosis granularity.(2)In the case of the item attribute pattern is partially missing in large-scale evaluation scenarios,the research is aimed at the problems that manually labeling the item attribute pattern is relatively difficult and subjective,and parametric modeling methods are relatively complex.(a)Constructing a cognitive diagnosis model based on supervised metric learning to predict objective item attribute pattern.(b)Constructing a cognitive diagnosis model based on unsupervised nearest neighbor to predict subjective item attribute pattern.(c)Using the above non-parametric models to predict and correct TestKnowledge-Ability tensor and 0/1 Q matrix of subjective and objective items respectively.Data evaluation proved that the method in this research makes the item labeling more objective and at the same time makes up for the shortcomings of the purely manual labeling method.In the prediction process,as long as experts mark a small part of the items,more accurate pattern results can be obtained,and the prediction and correction of attribute patterns of the rest items as well as the learners’ cognitive level and knowledge mastery can be automatically completed,which greatly saves the cost of human resources and time.At the same time,the empirical evaluation proves that the above method has high test quality,and its attribute match ratio is above 80%.The predicted item attribute pattern and the cognitive level of learners can provide accurate input information for learning diagnosis at the source,and promote the development of intelligent learning diagnosis technology.(3)Aiming at the problem that only a single factor affects learners is considered when personalizing the test paper based on the above learning analysis results,this research designs a personalized test paper generation strategy that integrates learners’ learning commonality,cognitive characteristics,prediction scores,and learning trends.Firstly,using the recommendation method based on knowledge point inverted index to obtain learning commonality.Secondly,the learner’s learning characteristics are obtained by cognitive diagnosis,and then use the probabilistic matrix factorization method to predict the learner’s potential scores on a specific item.Finally,providing a personalized test plan that increases learners’ knowledge mastery and is of moderate difficulty according to the test objectives.Experiments show that this research has a greater performance improvement than the traditional method,and the mean square error of the predicted score is below 0.14,and the question repetition rate is reduced to 30%,which ensures the security of the item bank and the interpretability of the test results is improved to a certain extent,and it can integrate richer and more reliable information to give learners more accurate and personalized test papers,thereby improving learners’ learning efficiency.This thesis studies the improvement of the key methods of intelligent learning diagnosis for intelligent evaluation by combining relevant knowledge in the field of computer,education,and psychology,perfecting the application research of intelligent learning diagnosis and enriching its use in personalized education,adaptive testing and other scenarios.