CRKT

Welcome to the official implementation of the paper "Enhancing Knowledge Tracing with Concept Map and Response Disentanglement" published in Knowledge-Based Systems (2024).

Paper Information

• Authors: Soonwook Park, Donghoon Lee, Hogun Park
• Journal: Knowledge-Based Systems
• Year: 2024
• DOI: 10.1016/j.knosys.2024.112346
• arXiv: 2408.12996

Citation

If you find our work useful, please consider citing it:

@article{park2024enhancing,
  title={Enhancing knowledge tracing with concept map and response disentanglement},
  author={Park, Soonwook and Lee, Donghoon and Park, Hogun},
  journal={Knowledge-Based Systems},
  pages={112346},
  year={2024},
  publisher={Elsevier}
}

Model Architecture

Experiment Environment

• python 3.10+
• torch 2.0+
• torch_geometric 2.4+
• scikit-learn 1.4+
• pandas 2.2.0+
• tqdm

Environment Setting Example

pip3 install torch torchvision torchaudio
pip3 install torch_geometric
pip3 install scikit-learn pandas tqdm

Dataset Download

To use the CRKT model, you first need to download and prepare the necessary datasets. The table below provides the download links and the required files for each dataset.

Dataset	Download	Required Files
DBE-KT22	link	Questions.csv, Question_KC_Relationships.csv, KCs.csv, Question_Choices.csv, Transaction.csv, KC_Relationships.csv
EdNet	link	kt_ednet.json
NIPS34	link	question_metadata_task_3_4.csv, train_task_3_4.csv, subject_metadata.csv
ENEM	link	enem_dep.csv

Setup Instructions

1. Download the datasets from the links provided in the table above.
2. Extract the files into the respective directories under data/{dataset_name}/raw. Ensure all required files are available in these directories before proceeding with preprocessing and experimentation.

Basic Usage

Preprocessing Datasets

Before running experiments with the CRKT model, you must preprocess the datasets. Use the following command to preprocess the dataset:

python3 data_preprocess.py --dataset {dataset_name}

Replace {dataset_name} with the name of the folder containing the dataset you want to preprocess. This should match the folder name within the data directory.

Run Experiment

To run an experiment with the CRKT model, use the following command:

python3 main.py --dataset {dataset_name} --model {model} --batch {batch_size} --lr {learning_rate} \
--dim_c {concept_dimension} --dim_q {question_dimension} --dim_g {gnn_output_dimension} \
--layer_g {gnn_layers} --lamb {lambda_value} --alpha {top_k_coefficient} --top_k {k_value} \
--beta {contrastive_loss_coefficient} --exp_name {experiment_name}

Arguments:

• --dataset {dataset_name}: The name of the dataset folder located in the data/ directory. Examples include DBE_KT22, EdNet, NIPS34, or ENEM.
• --model {model}: The name of the model you want to use for the experiment.
• --batch {batch_size}: The size of the batch used during training.
• --lr {learning_rate}: The learning rate for the optimizer.
• --dim_c {concept_dimension}: The dimension size for the concept embeddings.
• --dim_q {question_dimension}: The dimension size for the question embeddings.
• --dim_g {gnn_output_dimension}: The output dimension of the GNN (Graph Neural Network).
• --layer_g {gnn_layers}: The number of layers in the GNN.
• --lamb {lambda_value}: The lambda parameter used in the loss function.
• --top_k {k_value}: The value of k for the top-k selection in the model.
• --alpha {top_k_coefficient}: The coefficient for the top-k loss.
• --beta {contrastive_loss_coefficient}: The coefficient for the contrastive learning loss.
• --exp_name {experiment_name}: A name for the experiment, used for tracking and logging purposes.

Example Command:

To run an experiment with the DBE-KT22 dataset and the CRKT model with a batch size of 128, learning rate of 1e-3, and specified dimensions and coefficients, the command would look like:

python3 main.py --dataset DBE_KT22 --model CRKT --batch 128 --lr 1e-3 \
--dim_c 32 --dim_q 32 --dim_g 32 --layer_g 2 --lamb 0.1 --top_k 10 \
--alpha 0.1 --beta 0.01 --exp_name "test"