Recommender Systems have become crucial in the modern world, commonly guiding users towards relevant content or products, and having a large influence over the decisions of users and citizens. However, ensuring transparency and user trust in these systems remains a challenge; personalized explanations have emerged as a solution, offering justifications for recommendations. Among the existing approaches for generating personalized explanations, using visual content created by the users is one particularly promising option, showing a potential to maximize transparency and user trust. Existing models for explaining recommendations in this context face limitations: sustainability has been a critical concern, as they often require substantial computational resources, leading to significant carbon emissions comparable to the Recommender Systems where they would be integrated. Moreover, most models employ surrogate learning goals that do not align with the objective of ranking the most effective personalized explanations for a given recommendation, leading to a suboptimal learning process and larger model sizes. To address these limitations, we present BRIE, a novel model designed to tackle the existing challenges by adopting a more adequate learning goal based on Bayesian Pairwise Ranking, enabling it to achieve consistently superior performance than state-of-the-art models in six real-world datasets, while exhibiting remarkable efficiency, emitting up to 75% less CO2 during training and inference with a model up to 64 times smaller than previous approaches.

• The code in this repository has been tested with Python 3.10 and Cuda 11.8
• You can install all required packages with pip install -r requirements.txt
• This framework was executed in a dedicated Windows 10 Pro machine with an Intel Core i7-10700K CPU @ 3.80GHz, 16GB RAM, and an NVIDIA GeForce 2060 GPU Super.

• The six datasets are available for download at https://zenodo.org/record/5644892
• datasets should be placed in the data folder, conforming the following structure:

  data
  ├── barcelona
  │   ├── IMGREST
  │   │   ├── data_10+10
  │   │   │   ├── IMG_TRAIN
  │   │   │   ├── IMG_TEST
  │   │   │   ├── ...
  │   │   ├── original_take
  │   │   │   ├── ...
  │   
  ├── madrid
  │   ├── ...
  │
  ├── ...
  

• Pre-trained BRIE models are available for download at https://drive.google.com/drive/folders/1y5HlMk3tyQyW2nNEeaRBR5rpnVJwh5K3
• Models should be placed in the models folder, conforming the following structure:

  models
  ├── barcelona
  │   ├── BRIE
  │   │   ├── best_model.ckpt
  │   
  ├── madrid
  │   ├── ...
  │
  ├── ...
  

• To train a BRIE model, run python main.py --stage train --city CITY_NAME --model BRIE --max_epochs EPOCHS [--batch_size BATCH_SIZE] [--lr LR] [--dropout DROPOUT] -d DIMS --workers NUM_WORKERS [--early_stopping] [--no_validation]
• For example, to train BRIE with the hyperparameters used in the paper, run python main.py --city barcelona --model BRIE --max_epochs 15 --batch_size 2**14 --lr 0.001 --dropout 0.75 -d 64 --workers 4 --no_validation

• To test a BRIE model, run python main.py --stage test --city CITY_NAME --model MODEL_NAME... [--batch_size BATCH_SIZE] --workers NUM_WORKERS
• Multiple models can be tested at once by specifying their names separated by spaces after the --model argument
• For example, to test BRIE against ELVis, run python main.py --stage test --city barcelona --model BRIE ELVis --batch_size 2**14 --workers 4

• Below are the performance results obtained by BRIE in the six datasets used in the paper, compared to the state-of-the-art models ELVis and MF-ELVis, as well as two basic baselines.

• The results for ELVis and MF-ELVis were obtained by running the code provided by the authors of the original papers: MF-ELVis, ELVis

• Sustainability comparisons (training time and emissions, model size, and inference time and emissions) can be found in the paper.

• If you use this code or reference this model, please cite our paper:

  • APA:

    Paz-Ruza, J., Alonso-Betanzos, A., Guijarro-Berdiñas, B., Cancela, B., & Eiras-Franco, C. (2023). Sustainable Transparency in Recommender Systems: Bayesian Ranking of Images for Explainability. arXiv preprint arXiv:2308.01196.
    

  • Bibtex:

    @misc{pazruza2023sustainable,
        title={Sustainable Transparency in Recommender Systems: Bayesian Ranking of Images for Explainability}, 
        author={Jorge Paz-Ruza and Amparo Alonso-Betanzos and Berta Guijarro-Berdiñas and Brais Cancela and Carlos Eiras-Franco},
        year={2023},
        eprint={2308.01196},
        archivePrefix={arXiv},
        primaryClass={cs.IR}
    }