Robustness of Meta Matrix Factorization Against Decreasing Privacy Budgets

This repository includes python scripts and ipython-notebooks necessary for conducting experiments utilizing MetaMF and NoMetaMF in the setting of decreasing privacy budgets. The five utilized datasets, i.e., Douban [1], Hetrec-MovieLens [2], MovieLens 1M [3], Ciao [4] and Jester [5] are given in this repository. Additionally, we provide code for constructing and analyzing three user groups of these datasets with a low, medium and high number of ratings (available via Zenodo: https://doi.org/10.5281/zenodo.4031011).

Usage

To reproduce our results, the ipython-notebooks must be executed in the following order:

1. Initialize Folder Structure.ipynb: Sets up a hierarchy of folders for saving the experimental results.
2. data/jester/Generation.ipynb: Preprocessing of the Jester dataset utilized in our studies.
3. Identification of User Groups.ipynb: Identification of users with a low, medium or high number of ratings.
4. Train and Evaluate Models.ipynb: Train and evaluate our models (i.e., MetaMF and NoMetaMF) on the provided datasets and user groups.
5. Visualize Results.ipynb: Visualize results of our experiments.
6. Test Personalization and Collaboration.ipynb: Visualize the item embeddings and weights of the rating prediction models.

Furthermore, MetaMF.py includes the implementation of MetaMF and our extension NoMetaMF. However, MetaMF.py does not need to be run.

Requirements

• Python 3
• numpy
• pandas
• sklearn
• torch
• matplotlib

Contributors

• Peter Müllner, Know-Center GmbH, Graz, pmuellner [AT] know [MINUS] center [DOT] at (Contact)
• Dominik Kowald, Know-Center GmbH, Graz
• Elisabeth Lex, Graz University of Technology, Graz

References

[1] Hu, L., Sun, A., Liu, Y.: Your neighbors affect your ratings: on geographical neighborhood influence to rating prediction. In: SIGIR’14 (2014)

[2] Cantador, I., Brusilovsky, P., Kuflik, T.: Second international workshop on information heterogeneity and fusion in recommender systems (hetrec2011). In: RecSys’11(2011)

[3] Harper, F. M., Konstan, J. A.: The movielens datasets: History and context. ACM Transactions on Interactive Intelligent Systems (TIIS) 5(4), 1–19 (2015)

[4] Guo, G., Zhang, J., Thalmann, D., Yorke-Smith, N.: Etaf: An extended trust antecedents framework for trust prediction. In: ASONAM’14 (2014)

[5] Goldberg, K., Roeder, T., Gupta, D., Perkins, C.: Eigentaste: A constant time collaborative filtering algorithm. Information Retrieval 4(2), 133–151 (2001)