BloomNet: A Robust Transformer-Based Model for Bloom's Learning Outcome Classification

• Proposes BloomNet, a BERT-variant fine-tuned on CLOs. Finds that TF-IDF encoding outperforms CNN, RNN, and attention models on out-of-distribution examples — a counterintuitive result that validates TF-IDF as a strong baseline.

• CLO and questions alignment with Blooms taxonomy is an issue

  • Suggestion is to make a transformer based model named 'Bloom Net'

Relevance

• Justifies inclusion of TF-IDF as one of the three extraction methods. The finding that TF-IDF "outperforms numerous strong baselines" is a direct citation you can use to defend the TF-IDF branch of your CLO tool.