Personalization of Programming Education: An NLP-based Bi-dimensional(…)
summary
Automated pipeline classifies 106 Python exercises using a 318-term Python lexicon expanded to 715 forms, BERTopic with BERT embeddings and lexicon filtering, and five supervised models on 5,522 student data points; lexicon cuts TF-IDF features by 88%, raises adjusted Rand index from 0.21 to 0.33 and stabilizes topic count around eight clusters, while difficulty prediction tops at R^2 ≈ 0.41 (RMSE ≈ 0.05) but remains noisy and coherence metrics stay weak, so topic classification works reliably while difficulty estimation needs better ground truth.
AI-written summary · about
chapters (20)
- 0:00Introduction
- 0:26Scaling challenges
- 1:29Exercise classification limits
- 2:34NLP and lexicons for classification
- 3:37Research questions and approach
- 4:40Dataset and labeling
- 5:43Python lexicon building
- 6:49Topic classification with BERTopic
- 8:15Difficulty prediction
- 9:20Lexicon and feature results
- 10:03Topic classification metrics
- 11:48Topic coherence analysis
- 13:12Model performance
- 14:38Educator implications
- 15:44Future research
- 17:12Answering research questions
- 17:55Main contributions
- 18:16Q&A
- 19:03Student answer classification
- 20:48Closing remarks
related talks
01h 26mBagging to BERT- A tour of applied NLP
31mWhen humans teach machines
45mDecoding Topics: A Comparative Analysis of Python’s Leading Topic Modeling Libraries Using Climate C
01h 39mNatural Language Processing (NLP) in Python - From Zero to Hero
44mTeaching Computers ABCs: A Quick Intro to Natural Language Processing
17mNatural Language Processing in Automation
48mIntroduction to Natural Language Processing with Python
25mInventing Curriculum using Python and spaCy