Group 12

Requirements

1. Python version

Python3

2.Install Pytorch with GPU support(CUDA)

3.Install Jupyter Notebook

pip install jupyter

How to run?

A. Repeated Training Simulation

Run the Simulation.ipynb directly.

jupyter notebook Simulation.ipynb

B. Single Training Simulation

Please follow the steps below to modify some codes in Simulation.ipynb :

1.set the single_train variable to True
2.uncomment the code in "single training" cell in notebook.
3.Then run the jupyter notebook again

Result (Most of results are from Repeated Training)

1.Homogenization Effect

Repeated Training

Single Training

2.Calibrated Recommendation (Repeated Training)

Without Calibration

With Calibration

3.Item Popularity (Repeated Training)

References

[1] How algorithmic confounding in recommendation systems increases homogeneity and decreases utility (RecSys '18) DOI:https://doi.org/10.1145/3240323.3240370

[2] Calibrated recommendations (RecSys '18) DOI: https://doi.org/10.1145/3240323.3240372

[3] Neural Collaborative Filtering WWW '17: Proceedings of the 26th International Conference on World Wide WebApril 2017 Pages 173–182https://doi.org/10.1145/3038912.3052569

[4] Matrix Factorization Library : https://github.com/benfred/implicit

[5] Neural Collaborative Filtering Code: https://github.com/yihong-chen/neural-collaborative-filtering

[6] Calibrated Recommendations Tutorial: https://github.com/ethen8181/machine-learning

[7] Advances in Bias-aware Recommendation on the Web: https://github.com/biasinrecsys/wsdm2021

r09944035vsfu1 / team12_recommendation