mirrors/SimCIT

Gwhere

Guess Where You Go: Generative Next Point-of-Interest Recommendation in Amap View Demo · Report Bug · Request Feature

About The Project

Gwhere is an end-to-end generative recommendation framework for next point-of-interest (POI) prediction, designed to overcome the scalability limits of token-based retrieval and the lack of spatial-world understanding in LLMs. It introduces a contrastive item tokenization method that fuses multi-modal signals (text, image, spatial, and collaborative data) into compact, discriminative semantic identifiers (SIDs), enabling efficient large-scale generative retrieval. Built on these SIDs, a spatio-temporal LLM—pretrained on real-world mobility corpora and fine-tuned with a reinforcement learning algorithm (EAKTO)—aligns closely with user behavior.

Getting Started

This is an example of how you may give instructions on setting up your project locally. To get a local copy up and running follow these simple example steps.

Prerequisites

This is an example of how to list things you need to use the software and how to install them.

Clone the repository:
```
git clone repo_name
cd repo
```
Install dependencies:

Python 3.9+
PyTorch 2.2.0
requirements.txt
```
pip install -r requirements.txt
```

SID Generation

Training the Model

To start distributed training, use the following command:

./run_train.sh

Parameters

--state_dict_save_path: Directory for model outputs.

Testing the Model

Use the following command to start testing:

./run_infer.sh

LLM training

coming soon

License

Distributed under the project_license. See LICENSE for more information.

Contact

If you have any questions or encounter difficulties, we welcome you to contact ours via GitHub Issues. We are dedicated to supporting you in resolving issues related to sid generation, facilitating a robust and efficient setup for your system.

Citing this work

Please cite the following paper if you find our code helpful.

@misc{zhai2025cognitivealignedspatiotemporallargelanguage,
      title={Cognitive-Aligned Spatio-Temporal Large Language Models For Next Point-of-Interest Prediction}, 
      author={Penglong Zhai and Jie Li and Fanyi Di and Yue Liu and Yifang Yuan and Jie Huang and Peng Wu and Sicong Wang and Mingyang Yin and Tingting Hu and Yao Xu and Xin Li},
      year={2025},
      eprint={2510.14702},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2510.14702}, 
}