Abstract: To address the issue of inaccurate matching question difficulty to user abilities in online assessments,a dynamic question difficulty evaluation model based on user behavioral data was constructed.The evaluation model relies on a two-parameter logistic model and maximum likelihood estimation to obtain user ability parameters,introduces a response efficiency factor to describe differences in the response process,constructs a weighted average of user comprehensive abilities,and,based on a unified range of comprehensive abilities and global question difficulty values,utilizes the probability function calculation theory of the two-parameter logistic model to obtain the question difficulty level through probability,thus achieving dynamic estimation of question difficulty.Comparative experiment results show that the question difficulty level estimated by the constructed evaluation model is consistent with the actual question difficulty level.Ablation experiments show that after introducing the response efficiency factor,the absolute error of the evaluation model is significantly reduced,and the overall evaluation performance is significantly improved.

Key words: online assessment, comprehensive abilities, question difficulty, basic user behavior data, evaluation model