# GIRAM

**Repository Path**: mmmz2/GIRAM

## Basic Information

- **Project Name**: GIRAM
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-03-15
- **Last Updated**: 2026-03-16

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# GIRAM

Source code for ICDE 2026 paper **"Efficient Model-Agnostic Continual Learning for Next POI Recommendation"**.

Datasets and derived POI categories are available in the `./data` directory.

### Requirements

Make sure the following Python packages are installed:

- `torch`
- `torch_geometric`
- `torchsde`
- `networkx`
- `pandas`
- `numpy`

### Preprocessing
To prepare the datasets, run the preprocessing script:
```
sh pre.sh
```

### Pretraining
To train the base model on the initial data block (datasets: NYC, TKY, CA), run:
```
python main_Flashback.py --mode pretrain --dataset NYC
```

### Continual Updating
To perform continual learning with different update strategies, run:
```
python main_Flashback.py --mode memory --dataset NYC
```
Available modes:
- `memory`: Applies the proposed GIRAM method for continual learning.
- `finetune`: Incrementally fine-tunes the model using only new data.
- `retrain`: Retrains the model from scratch using all accumulated data.

### Acknowledgement
Our code is based on the following works:
- Flashback: https://github.com/eXascaleInfolab/Flashback_code
- GETNext: https://github.com/songyangme/GETNext
- DiffPOI: https://github.com/Yifang-Qin/Diff-POI

### Citation
Please kindly cite our work if you find our paper or codes helpful.

Link: https://arxiv.org/abs/2511.08941
```
@article{wang2025efficient,
  title={Efficient Model-Agnostic Continual Learning for Next POI Recommendation},
  author={Wang, Chenhao and Feng, Shanshan and Chen, Lisi and Li, Fan and Shang, Shuo},
  journal={arXiv preprint arXiv:2511.08941},
  year={2025}
}
```