Recommendation system evaluation for collaborative filtering, content-based & hybrid approaches

Thesis about recommendation systems. In this thesis a deep evaluation of a collaborative filtering method, conten-based method and hybrid approach has been carry out.

• For the Collaborative filtering method, a Matrix factorization approach was evaluated using implicit.py.
• For the Content-based method, a simple class for tf-idf recommendations was built using sklearn.
• The hybrid approach just combines the results of collaborative filtering and content based methods by mixing them.

Repository contents

• main.py: "Control panel" script for choosing options for the evaluation (which metrics, methods, randomize fold...)

• Plots/: Folder containing plots for dataset visualization.

• evaluate.py: Main loop of evaluation of the three methods.

• collaborative_filtering/:

• content_based

• data_visualization/: Scripts for reading the results and obtaining metrics.

• get_dataset.py: Script for download & extract the dataset.

• ReadSave.py: Simpler .pkl object read/saver.

• backup.pkl

• metrics.py: Ranking metrics implementations from this Gist.