Bloom's Learning Outcomes' Automatic Classification Using LSTM and Pretrained Word Embeddings

IEEE Access · 2021 · Shaikh, Daudpotta & Imran

• Authors are from Norway with one being from IBA Sukkur
• First deep-learning attempt at CLO Bloom-level classification.
• LSTM + pretrained word embeddings achieves 87% accuracy on CLOs and 74% on exam questions. Compares keyword/rule-based vs. ML approaches, finding rule-based works well for lower Bloom levels but fails on overlapping verbs.

Relevance

• Directly relevant to your CLO Extraction Tool's rule-based taxonomy method. The finding that rule-based keyword matching degrades on higher Bloom levels justifies adding your LLM layer. The 87% accuracy benchmark is the bar to beat or match.