Studying Retrieval Practice in an Intelligent Tutoring System
summary
Analysis of ALEKS (an adaptive ITS) uses a dataset of roughly 4 million students to extract ~850,000 assessment retrieval examples and ~300,000 learning retrieval examples, then fits logistic regression and multilevel controls to compare retention measured by randomized extra problems. Findings: waiting seven+ days for retrieval associates with better long‑term retention, and learning a post‑requisite (learning retrieval) yields about a 3.5 percentage‑point higher correct rate than assessment retrieval after controlling for confounders.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 0:22Problem background
- 1:45ALEKS system details
- 3:08Student workflow in ALEKS
- 4:51Retrieval practice setup
- 5:32Assessment retrieval example
- 6:14Learning retrieval example
- 6:56Exploratory analysis overview
- 7:36Forgetting curves results
- 9:42Learning retrieval curves
- 10:25Direct comparison
- 11:06Statistical modeling
- 12:31Possible explanations
- 13:11Discussion and summary
related talks
15mAdaptive, AI-Driven Security Assessments
01h 30mOptimizing Retrieval for Agentic Systems
01h 05mAdaptive Generalization: The Role of Dynamic Evaluation in Low-Resource Settings
33mHow AI Adaptive Technology can upgrade Education
51mEfficient Retrieval of Influential LLM Training Examples
12mAI-Driven Assessments for Continuous Learning
29mRemember What You Learn (Spaced Repetition Model)
23mKeynote: What is Learning in the AI Era?