TelaRec

Building Sequential Recommendation Models upon Thermoelastic Spectral Operators (CIKM 2026 Full Research Paper)

📊 Benchmark Datasets

This project evaluated the TelaRec model on the following benchmark datasets:

• Amazon_ratings (Rating-only categories):
  • Beauty
  • Sports & Outdoors
  • Video Games
  • Electronics
• Gowalla (Location-based social network dataset):
  • Gowalla-Merged
• MovieLens (Popular movie recommendation benchmark dataset):
  • MovieLens-1M

Dataset Download & Placement

You can obtain the preprocessed atomic files (.inter) for these datasets using the following methods:

1. Automatic Download

RecBole supports automatic downloading for several standard benchmark datasets. When you run a command using these datasets for the first time, they will be automatically downloaded and extracted into the dataset/ directory.

2. Manual Download from Official Channels

You can manually download the preprocessed atomic files from the following official channels:

• GitHub Repository: RUCAIBox/RecDatasets
• Google Drive: Processed Datasets in Google Drive
• Baidu Wangpan: Baidu Wangpan Link (Extraction Code / Password: e272)

Tip

If you download the datasets manually, please ensure they are extracted and placed under the dataset/ folder in the project root (e.g., dataset/beauty/, dataset/gowalla-m/) so the program can locate them correctly.

---

🚀 How to Run TelaRec

To train and evaluate the TelaRec model on a specific dataset, execute the run_recbole.py script from the project root directory:

python run_recbole.py --model=TelaRec --dataset=<dataset_name>

Examples

• Run on Amazon Beauty dataset:

  python run_recbole.py --model=TelaRec --dataset=beauty

• Run with custom hyperparameters (e.g., modifying learning rate and embedding size):

  python run_recbole.py --model=TelaRec --dataset=beauty --learning_rate=0.001 --embedding_size=64

⚙️ Configuration & Optimal Hyperparameters

The default hyperparameters for TelaRec are stored in the configuration file:

• Configuration File Path: TelaRec.yaml

Optimal Hyperparameters for Benchmark Datasets

Here are the optimal settings for c_init, alpha_init, and kappa_init extracted from TelaRec.yaml. You can run them by passing the hyperparameters directly in the command:

Dataset	c_init	alpha_init	kappa_init	Example Run Command
Beauty	1.0	0.001	0.1	python run_recbole.py --model=TelaRec --dataset=beauty --c_init=1.0 --alpha_init=0.001 --kappa_init=0.1
Sports	0.5	0.0001	0.5	python run_recbole.py --model=TelaRec --dataset=sports --c_init=0.5 --alpha_init=0.0001 --kappa_init=0.5
Video	0.1	0.001	0.5	python run_recbole.py --model=TelaRec --dataset=video --c_init=0.1 --alpha_init=0.001 --kappa_init=0.5
Elec	0.5	0.01	0.01	python run_recbole.py --model=TelaRec --dataset=elec --c_init=0.5 --alpha_init=0.01 --kappa_init=0.01
Gowalla-M	0.1	0.01	0.001	python run_recbole.py --model=TelaRec --dataset=gowalla-m --c_init=0.1 --alpha_init=0.01 --kappa_init=0.001
ML-1M	1.0	0.001	0.001	python run_recbole.py --model=TelaRec --dataset=ml-1m --c_init=1.0 --alpha_init=0.001 --kappa_init=0.001

Acknowledgments

This implementation is based on the RecBole recommendation library. We appreciate their outstanding work.