ahmadarialdino/lcf-bert-classifier

Learner-Centered Feedback Classifier (BERT)

A BERT-based, multi-label classifier that detects learner-centered feedback (LCF) components in written feedback, from:

Aldino, A. A., Tsai, Y.-S., Mello, R. F., Gašević, D., & Chen, G. (2024). Enhancing feedback quality at scale: Leveraging machine learning for learner-centered feedback. Computers and Education: Artificial Intelligence, 7, 100332. https://doi.org/10.1016/j.caeai.2024.100332

• Paper: https://www.sciencedirect.com/science/article/pii/S2666920X24001358
• App and code: https://github.com/ahmadarialdino/learner-centered-feedback-classifier

Labels

The model outputs 8 sigmoid probabilities, in this order:

Index	Label	Component
0	A_Active_Role	Active Role
1	A_Affirm_Encourage	Affirmation and Encouragement
2	A_Strengthen_Relationship	Student–Teacher Relationship
3	FI_Combination	Meeting Learning Objective
4	FI_Future_Improvement	Upcoming Similar Tasks
5	Non_LCF	Non-LCF
6	S_Performance_Summary	Performance Summary
7	S_Strengths_Weaknesses	Strengths and Weaknesses

Usage

Feedback is split into sentences, each sentence is classified (max length 100 tokens), and a component counts as present in the feedback if any sentence scores at least 0.50.

import torch
from transformers import AutoModelForSequenceClassification, AutoTokenizer

repo = "ahmadarialdino/lcf-bert-classifier"
tokenizer = AutoTokenizer.from_pretrained(repo)
model = AutoModelForSequenceClassification.from_pretrained(repo).eval()

sentences = ["Your literature review is a real strength.", "Next time, link each finding to the research question."]
with torch.no_grad():
    probs = torch.sigmoid(model(**tokenizer(sentences, padding=True, truncation=True, max_length=100, return_tensors="pt")).logits)
print(probs >= 0.5)

Predictions are model-assisted estimates and should support, not replace, human judgement.

Citation

@article{aldino2024enhancing,
  title   = {Enhancing Feedback Quality at Scale: Leveraging Machine Learning for Learner-Centered Feedback},
  author  = {Aldino, Ahmad Ari and Tsai, Yi-Shan and Mello, Rafael Ferreira and Ga{\v{s}}evi{\'c}, Dragan and Chen, Guanliang},
  journal = {Computers and Education: Artificial Intelligence},
  volume  = {7},
  pages   = {100332},
  year    = {2024},
  doi     = {10.1016/j.caeai.2024.100332},
  url     = {https://www.sciencedirect.com/science/article/pii/S2666920X24001358}
}