CuckooSearch-Kmeans Algorithm

Made by me and my estimated collegue Ortensia Forni
CuckooSearch-Kmeans algorithm (CS-Kmeans) is a learning framework to make recommendations. The key idea is to learn the user-item interaction using matrix factorization (MF) or neural collaborative filtering (NCF) to project them into a latent space and explore the latter to find the best clusterization to use as input a model-based collaborative filtering algorithm.

Dataset

ratings.csv is used to train and test the model and movies.csv to get titles of recommended movies.

Files

project.pdf: contains full project pdf, made for an university project.

methods.py: contains the cuckoo search, k-means, cuckoo-kmeans and relative cross validation

utils.py: some handy functions for model training etc.

nnls.py and NonNegativeMatrixFactorization.py: Block Principal Pivoting Method and Non Negative Matrix Factorization

function.py: objective function environment

run_examples.ipynb: runs cuckooo search on some test functions as Ackley and Himmelbau

run_mf.ipynb: runs matrix factorization (MF), cross validating it and producing user_latent_matrix_mf.py

run_ncf.ipynb: runs neural collaborative filtering (NCF), cross validating it and producing user_latent_matrix_ncf.py

user_latent_matrix_mf.py and user_latent_matrix_ncf.py: user latent matrices produced respectively by MF and NCF methods

main.ipynb: runs the CS-Kmeans algorithm implemented with both MF and NCF and tests their accuracies