CONTRAST: Contrastive Temperature-Regulated Adaptive Session Training

This is the official implementation for the "Adaptive Session-based Recommendation with Contrastive Learning" paper. It includes data processing of datasets (RetailRocket and Diginetica) that we used to train and test our model.

🔗 Notebooks on Kaggle

• CONTRAST Data Pipeline: Clean, transform, and structure raw data into usable session datasets.
• CONTRAST Model Implementation: Architecture with training, evaluation, and visualization.

---

Datasets

The framework was evaluated on two benchmark datasets:

Dataset	Train Sessions	Test Sessions	Unique Items	Avg. Session Length
RetailRocket	433,643	15,132	41,762	5.64
Diginetica	719,470	60,858	37,867	4.06

The preprocessing steps are encapsulated in the Data Pipeline Notebook.

---

Performance Table

---

Running the Notebooks

To experiment with the CONTRAST model or its data pipeline:

1. Open the corresponding Kaggle notebook links.
2. Fork the notebook to your Kaggle account.
3. Run all cells (ensure GPU is enabled for faster training).

---

License

Licensed under the Apache License.