ContextsPOI: Contextual Information on POI Recommendation

How to run the framework?

All of the codes to run the models and processing are located in codes folder.

1. At first step, we need to run the contextual models to get the preference scores of users on POIs based on each contextual influence. To do this, we provide all the contextual models in the contextsModels package which are in lib. When you run saveScores.py, all contextual models which are located in lib compute the users' preference score, then they will be saved as NumPy arrays (.npy) in /model_combiner/contexts/dataset and the dataset can be Yelp or Gowalla, indicates on which dataset your run the models and get these results. The size of these produces NumPy arrays is equal to user_num * poi_num.
2. The second step is the embedding of users and POIs that we need them as for the embedding layer of user and POIs in Neural Network based approaches. To embed the user and location into the latent feature space, you can run the embedding.py from the MFEmbedder package. the output is two Numpy arrays U.npy and L.npy which are saved in /embeddings/dataset/.
3. Run the base models (MF and NN) to achive results on linear and non-linear models. To do this, you can run main.py in codes and save the results automatically to Numpy arrays. All of this process will be done automatically with code, for example, for Neural Network based models the code first loads the embedding vectors of users and items.
4. Combine all possible combination of MF and NN models with contextual models separately. In order to combine differenc contextual models with each other and also linear and non-linear models we can use the code on the model_cobiner package. What we need to do is to config the model to evaluate models and context. In model_combiner.py we can define our specific configuration, we can select the dataset, base models, contextual models. Then, the results will be written under the results folder.

In the next version of our codes, we are going to make it easier to run, using the command-line arguments.

Prerequisites

You will need below libraries to be installed before running the application:

• Python >= 3.8
• Numpy >= 1.17.3
• Tensorflow >= 2.2.0
• SciPy >= 1.4.1

For a simple solution, you can simply run the below command in the root directory:

pip install -r prerequisites.txt

Citation

If you find ContextsPOI useful for your research or development, please cite the following paper:

@article{rahmani2022context,
  title={A Systematic Analysis on the Impact of Contextual Information on Point-of-Interest Recommendation},
  author={Rahmani, Hossein A. and Aliannejadi, Mohammad and Baratchi, Mitra and Crestani, Fabio},
  journal={ACM Transactions on Information Systems (TOIS)},
  volume={-},
  number={-},
  pages={--},
  year={2022},
  publisher={ACM New York, NY}
}

Team

• Hossein A. Rahmani, Web Intelligence Group, University College London, United Kingdom (h.rahmnai@ucl.ac.uk)
• Mohammad Aliannejadi, IRLab, University of Amsterdam, The Netherlands (m.aliannejadi@uva.nl)
• Mitra Baratchi, ADA Research Group, Leiden University, The Netherlands (m.baratchi@liacs.leidenuniv.nl)
• Fabio Crestani, IR-USI, Università della Svizzera italiana (USI), Switzerland (fabio.crestani@usi.ch)

Contact

If you have any questions, do not hesitate to contact us at rahmanidashti@gmail.com, we will be happy to assist.

Acknowledgements

This work was in part supported by the Swiss State Secretariat for Education, Research and Innovation (SERI) mobility grant between Switzerland and Iran, and in part by the NWO (No. 016.Vidi.189.039 and No. 314-99-301).