CS4248_project - Group 04 : Instructor Intervention Prediction In MOOC

1. Environment setup

• python version = 3.6.12

• pip install -r requirements.txt

2. Dataset

Training corpora are from Task1-Marking_Task in NUS-MOOC-Transacts-Corpus

3. Run program

Preprocessing and feature engineering over the corpora takes long time (nearly 2 hours), so we save feature engineering vectors into files, then we only need to load these files for next studying such as : hyper-parameter tuning, ablation testing, model perfomance verifying ...

3.1 tf_idf feature:

a. load raw files from dataset and process feature engineering :

- python main.py --feature tf_idf --dataset unprocessed

b. load feature engineering dataset (processed in step a):

- python main.py --feature tf_idf --dataset processed

3.2 word2vec feature:

a. load raw files from dataset and process feature engineering :

- python main.py --feature word2vec --dataset unprocessed

b. load feature engineering dataset (processed in step a):

- python main.py --feature word2vec --dataset processed

4. Model performance

4.1 tf_idf feature:

- f1 score on test data = 0.88
- accuracy score on test data = 0.81
- precision score on test data = 0.85
- recall score on test data = 0.92
- f2 score on test data = 0.91

4.2 word2vec feature:

- f1 score on test data = 0.88
- accuracy score on test data = 0.82
- precision score on test data = 0.85
- recall score on test data = 0.92
- f2 score on test data = 0.90