maximelel / ml4ed_ssl_approaches

Semi-Supervised Learning Approaches in Educational Contexts

Semi-Supervised Learning Approaches in Educational Contexts

Semester Project in the ML4ED (Machine Learning for Education) lab at EPFL (Sept. 2023 - Jan. 2024)

🌐 Overview

📚 Context: Reflective Writings of student teachers.

• Educational Benefits of Reflective Writing: Reflection on one’s performance through reflective writing offers substantial educational benefits.

• Challenges in Reflective Writing: Students need a very personnalized feedback from the educators and educators face complexities in teaching and assessing reflective writings.

• ML and NLP Solutions: ML and NLP offer potential solutions to challenges in reflective writing by automating the feedback provision.

🚧 Challenge

• Labeling Hurdles in Education: Grapple with the challenges of labeling large datasets in the educational context, demanding expert involvement and high pedagogical knowledge.

✨ Solution Proposed

• Semi-Supervised Learning with Self-Training: Explore the potential of self-learning, a semi-supervised approach that overcomes labeling challenges by leveraging small amounts of labeled data in multi-class text classification, specifically in the nuanced analysis of reflective writings.

🔍 Research Question

Can semi-supervised approaches reduce human effort in labeling reflective writings?

📊 Methodology

• Baseline with BERT Models: Establish baselines for multi-class and binary classification on labeled reflection datasets.

• Learning Curves Analysis: Explore the influence of dataset size on model performance and confidence.

Reproducibility

File structure

• EDA_notebook : notebook for the exploratory data analysis of the CeRED dataset.
• notebook_kaggle : notebook used to experiment and test the functions before running the code on the cluster
• run_to_cluster : directory with all the files necessary to run the code on the cluster (see README for more information).
• results_visualization : directory containing 2 jupyter notebooks to visualize the results (saved in the directory results txt files) after running the code on the cluster.
• results txt files : results saved on txt files. Can be directly reused for visualization in the notebooks in results_visualization
  • Multi_Class_CLF
  • Multiple_Bin_CLF

Report

The full report of this project can be found here.