The authors labeled all 1352 LectureBank lectures using TutorialBank taxonomy for classification. There are 305 topics in total.

University of Michigan - Ann Arbor

- Data collection and preprocessing
- Prerequisite annotation
- Classification
- Vocabulary

What Should I Learn First: Introducing LectureBank for NLP Education and Prerequisite Chain Learning: Dataset

Fabbri, A. R., Li, I., Trairatvorakul, P., He, Y., Ting, W. T., Tung, R., ... & Radev, D. R. (2018). Tutorialbank: A manually-collected corpus for prerequisite chains, survey extraction and resource recommendation. arXiv preprint arXiv:1805.04617.

https://arxiv.org/pdf/1805.04617.pdf

Improved Concept Embeddings for Learning Prerequisite Chains: References for TutorialBank

- LectureBank includes online lecture files, containing 60 courses in 5 different domains, NLP, Machine Learning, AI, deep learning, and information retrieval. 
- All the files are .pdf and/or .ppt(x). PDFMiner and python-pptx packages are used to extract text from the files. Keep the .pdf file if both versions are provided.

What Should I Learn First: Introducing LectureBank for NLP Education and Prerequisite Chain Learning: Data collection and preprocessing

- The authors chose binary annotation that only have answers of yes and no, implying that even if the distance between two concepts in a concept graph is long, as long as "one concept is typically learned before the other" and "is helpful in building up knowledge" for learning the other, the author considered that a prerequisite relation exists.
- Based on the concept labels provided by TutorialBank (Fabbri et al. 2018), the annotators constructed a directed concept graph with 208 vertices (concept) and 921 edges (A --> B means A is a prerequisite of B).
- In the concept graph, cycles (a pair of very closely related concepts) and independent vertices (no prerequisite relations with any other concepts) exists.
- In-degree: the number of the concept's prerequisite vertices; high in-degree means the concept is an advanced concept.
- Out-degree: the number of vertices that the concept is prerequisite to; high out-degree means the concept is a fundamental concept.
- The longest path consists of 14 concepts.

What Should I Learn First: Introducing LectureBank for NLP Education and Prerequisite Chain Learning: Prerequisite annotation

What Should I Learn First: Introducing LectureBank for NLP Education and Prerequisite Chain Learning: Classification

The author combined three topic sets:
1. the TutorialBank taxonomy topics (305);
2. the TutorialBank concepts labeled for prerequisite relations (208);
3. the LectureBank topics that automatically extracted from the header section of each lecture slide.

There are 1221 terms in total.

Learn Before

Related