Knowledge Tracing Transformer (SQKT)

Setup

Requirements

We recommend using Python 3.8+ and creating a virtual environment.

Install the dependencies via pip:

pip install -r requirements.txt

---

Data Format

The model requires four types of CSV files per dataset. These are usually located under a folder like data/.

1. exercises.csv

Houses instructional text and solution code blocks for each problem.

Column	Description
exercise_id	Unique problem identifier
Instruction Content	Problem description
Solution Content	Reference solution text

2. submissions.csv

Contains student submissions, their code, and timestamps.

Column	Description
x_user_id	Unique student identifier
exercise_id	Problem ID
contents	Submitted code text
created_datetime	Submission time (Unix timestamp in ms)

3. questions.csv

Logs of student-educator interactions (both student questions and educator responses and its timestamp).

Column	Description
x_user_id	Student ID
exercise_id	Problem associated with the question
content	Text of the question or answer
post_created_datetime	When the entry was posted
is_student	True for student, False for teacher reply

4. scores.csv

Label data indicating whether a student solved a problem correctly.

Column	Description
x_user_id	Student ID
exercise_id	Problem ID
label	Binary label (0 or 1)

---

Training

python main.py 

---

Metrics

Evaluation is done using:

• Accuracy
• Precision
• Recall
• F1 Score
• AUC (ROC)

Sample output after each epoch:

Epoch 1/10 completed, Average Train Loss: 0.4593
Validation Metrics - Accuracy: 0.81, Precision: 0.79, Recall: 0.83, F1 Score: 0.81, AUC: 0.86

---

Contact

[email protected]