POIGDE

This is the pytorch implementation for paper:

Siamese Learning based on Graph Differential Equation for Next-POI Recommendation.

Please cite our paper if you use the code.

Environment Requirement

The code has been tested running under Python 3.8.10. The required packages are as follows:

• pytorch == 1.11.0
• torch-geometric == 2.1.0
• pandas == 1.5.1
• hydra-core == 1.3.2
• torchsort == 0.1.9

Running

Here is the process of running the model with NYC dataset.

1. unzip dataset

In the main folder, unzip the data package.

unzip data.zip

data.zip contains the dataset of NYC. You can download dataset of TKY and SG from here .

We filter out the user with interaction times less than 10 and POI visited times less than 10.

2. create new folder

In the main folder, create a folder to store running records and another folder to store checkpoint.

mkdir logs
mkdir ckpts

3. preprocess the dataset

Run process.py to generate data for model from the dataset (operate on NYC by default).

python utils/process.py

4. run the model

Run the model with the following command.

python ode_main.py hydra.job.chdir=False