Student Ability Evaluation And Achievement Prediction Based On IRT And Transformer

In recent years,adaptive learning based on AI technology has played an important role in improving learning efficiency,reducing learning time and promoting educational equity,and has become a research hotspot in the field of education.In the field of adaptive education,how to evaluate students’ ability and how to predict students’ performance is still the focus of current research.The main research contents of this paper are as follows:（1）Aiming at the problem of less test data in the learning process,this paper analyzes the parameters B and a in the logistic model,and puts forward an estimation method of item response theory（IRT）in the case of small samples.For parameter B,this paper explores the strong correlation between the estimation results of classical test theory（CTT）and IRT,and puts forward the CTT estimation method for the difficulty parameter when the tested capacity distribution is non-standard normal distribution.For parameter a,by exploring the impact of parameter a on the estimation effect in the estimation process,this paper puts forward the method of taking the fixed parameter as the value of parameter a,The results show that this method reduces the amount of calculation,shortens the estimation time,and has a good estimation effect.（2）Aiming at the problems of poor interpretability and lack of learning characteristics in the field of performance prediction in deep knowledge tracking（DKT）,this paper proposes an improved transformer structure performance prediction model based on learning process.The coding layer of the model is divided into three coding networks,all of which adopt the multi head self attention mechanism.The inputs are preprocessed three types of data: process feature data,test label data and answer result data,and the outputs are V,Q and K matrices;The decoding layer also adopts the attention mechanism.The input is the three matrices of the coding layer,and the output is the prediction result.The V,Q and K matrices in the attention mechanism are explicable to a certain extent: the V matrix represents the characteristic information of the subjects’ learning process,the Q matrix reflects the knowledge point information investigated by the current test topic,and the K matrix represents the result information of the previous test,thus reflecting the causal relationship between the learning process and the test results to a certain extent.In this paper,three datasets,assist2017,ednet and static2011,are used for experimental verification.The results show that its ACC and AUC evaluation indicators have been improved on ednet and assist datasets,and have also achieved good performance on static datasets.