The Unfairness of Active Users and Popularity Bias in Point-of-Interest Recommendation

This project focuses on the point of interest (POI) recommendation and state-of-the-art algorithms in the POI domain regarding the trade-offs between the accuracy of personalizing and fairness in recommendations to both users and providers.

💡 Evaluation Results

By analyzing the performance of various algorithms on three real-world POI datasets, we found that:

• State-of-the-art algorithms work against user fairness by favoring a small percentage of highly active users with superior recommendations.
• A more significant percentage of less active users receive imprecisely tailored recommendations.
• Most recommender systems recommend popular locations or items contributing to item unfairness.

⚠️ Note: Visit our web page for full table results.

☑️ Re-generating the results

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

• Python >= 3.7
• NumPy >= 1.19
• SciPy >= 1.3

You can also run the command below in the root directory to get all of them installed:

pip install -r requirements.txt

⚙️ Team

We are a diverse group of individuals who bring perspectives to the state-of-the-art projects:

Yashar Deldjoo	Hossein A. Rahmani	Ali Tourani	Mohammadmehdi Naghiaei

📝 Citation

Please cite the following paper:

@misc{rahmani2022unfairness,
      title={The Unfairness of Active Users and Popularity Bias in Point-of-Interest Recommendation},
      author={Hossein A. Rahmani and Yashar Deldjoo and Ali Tourani and Mohammadmehdi Naghiaei},
      year={2022},
      eprint={2202.13307},
      archivePrefix={arXiv},
      primaryClass={cs.IR}
}