Their model outperformed every model on each of these three dataset:



San Diego State University

1. Experimental Setting
2. Results
3. Discussion

Experiments (DAS3H: Modeling Student Learning and Forgetting for Optimally Scheduling Distributed Practice of Skills)

To evaluate the DAS3H model, researchers split the student population into five disjoint groups for cross-validation. Every feature weight and embedding component was set to follow a normal prior distribution, and the models were implemented in Python. The DAS3H model was compared against the DASH, IRT, PFA, and AFM models using 0, 5, and 20 embedding dimensions. These experiments utilized three datasets: ASSISTments (assist12), Bridge to Algebra (bridge06), and Algebra I (algebra05).

Experimental Setting (DAS3H: Modeling Student Learning and Forgetting for Optimally Scheduling Distributed Practice of Skills)

Results (DAS3H: Modeling Student Learning and Forgetting for Optimally Scheduling Distributed Practice of Skills)

The impact of the embedding dimension on the performance of the DAS3H model is variable. Across evaluated datasets, setting the embedding dimension to 20 consistently resulted in lower performance, while a dimension of 5 sometimes yielded better performance than a dimension of 0. Furthermore, using time window features—rather than regular skill wins and attempt counts—alongside distinct parameters for individual skills in $$h_{\theta}$$ increased the model's Area Under the Curve (AUC). Incorporating both item-skill relationships and the forgetting effect improves performance compared to models that consider only one of these factors, making DAS3H well-suited for adaptively optimizing personalized skill practice schedules.

Learn Before

Related