Paper-Implementation-Matrix-Factorization-Recommender-Systems-Netflix

IEEE paper "Matrix Factorization Techniques for Recommender Systems"

- Yehuda Koren, Robert Bell, Chris Volinsky

Python 3.6

Links to original paper published by IEEE Computer Society : [1], [2]

Link to Netflix Dataset Used : [1]

Files

1. Presentation.pdf : Explains the paper. Was written in Latex Beamer, tex code is in presentation.tex

2. recommender_final.py : The final recommender. Includes biases and regularization. Requires mf.py to be imported to run. Use directly on any dataset by changing line 19 in recommender_final.py.

3. recommender_final_toy_dataset.py shows how exactly Matrix Factorization Techniques work by considering a 5x5 toy dataset.

4. The .ipynb_ files include visualizations of RMSE decreasing with iterations when fitting on the training dataset. All .ipynb files are standalone and do not require importing mf.py

5. feasible_data_n.txt : Files with only the first n datapoints from whole dataset. Used for Testing.

6. Training and Testing Data :
   Not given separately. Program randomly separates k% of data as Test data, trains on remaining, then tests on the k% values. Default k=20, can be changed on line 154.

Error Analysis