Data-driven internal mobility: Similarity regularization gets the job done
_{Simon De Vos, Johannes Desmedt, Marijke Verbruggen, Wouter Verbeke [2024]}

Instructions:

To replicate the results as reported in the paper, exectute the notebook 'main.ipynb'

• Instructions for 'main.ipynb':
  • Set the project directory to your custom folder. For example, DIR = r'C:\Users\folder\subfolder\RecSys_SR'
  • Specify experiment configuration in settings = {'dataset': 3, 'method': 'mfsr', ...}. The settings as currently specified will replicate the results on dataset 3. Per key, the following options are available:
    • dataset: {3}, 1 and 2 are not publicly available. Select 0 for a small synthetic dataset for quick runs and debugging.
    • method: {'mf', 'mfsr', 'svd', 'slopeone', 'knnc', 'knnp', 'knnsmd', 'knnta'},
    • test_size: [0,1],
    • oot: {0,1}, out-of-time split (no/yes)
    • oitv: {0,1}, ensures one observation per entity in the training set (no/yes)
    • load_similarity: {False, True}, If True, load similarity matrix from file, otherwise compute it. Relevant for methods: 'mfsr', 'knnsmd', 'knnta'
    • hyperpara_tune: {False, True}, If True, tune hyperparameters. If False, load optimal hyperparameters from 'optimal_hyperparameters.yaml'
  • Datasets 1 and 2 are not publicly available. Dataset 3 can be found here.
  • Hyperparameter grids can be adapted in 'hyperparameter_grid.yaml'

Project structure:

RecSys_SR/
│
├── data/
│   ├── toy_example.csv
│   └── dataset_3.csv
│
├── experiment/
│   ├── experiment.py: defines Experiment class and functions
│   ├── hyperparameter_grid.yaml: defines hyperparameter grid
│   ├── main.ipynb: run this to replicate experiments
│   ├── methods.py: defines methods ('mf', 'mfsr', 'svd', 'slopeone', 'knnc', 'knnp', 'knnsmd', 'knnta')
│   └── utils.py: contains utility functions
│
├── img/
│   ├── Two images on latent representations (Figures 2a and 2b).
│   └── EJM (Figure A.3)
│
└── similarity_matrix/
    └── stored_similarity_matrices for methods 'mfsr', 'knnsmd', 'knnta'

Citing

Please cite our paper and/or code as follows:

De Vos, S., De Smedt, J., Verbruggen, M., & Verbeke, W. (2024). Data-driven internal mobility: Similarity regularization gets the job done. Knowledge-Based Systems, 111824.