Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Recurso educativo Open Access |
| Lenguaje: | en |
| Publicado: |
2023
|
| Materias: | |
| Acceso en línea: | https://eric.ed.gov/?id=ED630880 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
Tabla de Contenidos:
- Meta-Learning for Better Learning: Using Meta-Learning Methods to Automatically Label Exam Questions with Detailed Learning Objectives Zur, Amir Applebaum, Isaac Nardo, Jocelyn Elizabeth DeWeese, Dory Sundrani, Sameer Salehi, Shima Equal Education Prior Learning Educational Objectives Chemistry Physics Textbooks Benchmarking Science Instruction Feedback (Response) Computer Assisted Testing Classification Networks Computational Linguistics Test Items Detailed learning objectives foster an effective and equitable learning environment by clarifying what instructors expect students to learn, rather than requiring students to use prior knowledge to infer these expectations. When questions are labeled with relevant learning goals, students understand which skills are tested by those questions. Labeling also helps instructors provide personalized feedback based on the learning objectives each student struggles to master. However, developing detailed learning objectives is time-consuming, making many instructors unable to pursue it. Labeling course questions with learning objectives can be even more time-intensive. To address this challenge, we develop a benchmark for automatically labeling questions with learning objectives. The benchmark comprises 4,875 questions and 1,267 expert-verified learning objectives from college physics and chemistry textbooks. This dataset provides a large library of learning objectives, and, to the best of our knowledge, is the first benchmark to measure performance on labeling questions with learning objectives. We use meta-learning methods to train classifiers and test them against our benchmark in a few-shot classification setting. These classifiers achieve acceptable performance on a test set with previously unseen questions (AUC 0.84), as well as a course with previously unseen questions and unseen learning objectives (AUC 0.84). Our work facilitates labeling questions with learning objectives to help instructors provide better feedback and create equitable learning environments. [For the complete proceedings, see ED630829.]