CTRA-Net Datasets

ASSISTments2009 is derived from an online mathematics learning platform for middle school students in the United States. It contains interaction records of 4,151 students across 110 problems, totaling 325,637 interactions and involving 110 knowledge components. The data are organized as time-ordered question–answer pairs, with each record containing the problem ID, associated knowledge component, and a binary correctness label.Statics2011 comes from a university-level engineering course. It includes 189,297 interactions from 333 students across 1,223 problems, covering 85 knowledge components. The dataset is of moderate scale with relatively concentrated knowledge distribution.Synthetic-5 is a simulated dataset containing interactions of 4,000 virtual students over 50 problems, totaling 200,000 interactions with 5 knowledge components. It is used to analyze the model’s behavior under simplified structural conditions.KDDCup2010 originates from an algebra tutoring system. It includes 607,026 interaction records from 575 students across 436 problems, involving 112 knowledge components. The interaction sequences are relatively long, making it suitable for evaluating stability in long-sequence scenarios.Eedi is a large-scale online mathematics learning dataset, containing 15,867,850 interactions from 118,971 students over 27,613 problems, involving 388 knowledge components. Its large scale and complex distribution make it appropriate for testing generalization performance in realistic application settings.DBE-KT22 is collected from a university-level database course. It includes 167,222 interactions from 1,361 students across 212 problems, covering 98 knowledge components. The dataset has a well-defined course structure, facilitating analysis of structural modeling effects.