Program implementation of a recommender system

U okviru ovog rada bavimo se implementacijom i vrednovanjem različitih metoda preporučivanja. Poseban naglasak stavljamo na metode neuronskog suradničkog filtriranja (NCF) i matrične faktorizacije (MF). U programskom okviru PyTorch ostvarujemo tri modela neuronskog suradničkog filtriranja: generaliziranu matričnu faktorizaciju (GMF), višeslojni perceptron (MLP) i neuronsku matričnu faktorizaciju (NeuMF). Dodatno, ostvarujemo model obične matrične faktorizacije i matrične faktorizacije s pomakom. Sve modele treniramo i evaluiramo na javno dostupnom skupu podataka MovieLens 1M. Provodimo različite eksperimente te uspoređujemo performanse modela na metrikama HR@k, NDCG@k i MRR@k. Svoja opažanja uspoređujemo s rezultatima iz literature.In this thesis, we focus on the implementation and evaluation of various recommendation methods. We place special emphasis on the methods of Neural Collaborative Filtering (NCF) and Matrix Factorization (MF). Within the PyTorch framework, we implement three NCF models: Generalized Matrix Factorization (GMF), Multi-Layer Perceptron (MLP), and Neural Matrix Factorization (NeuMF). Additionally, we implement a model of plain matrix factorization and matrix factorization with bias. All models are trained and evaluated on the publicly available MovieLens 1M dataset. We conduct various experiments and compare the model's performances using metrics such as HR@k, NDCG@k, and MRR@k. We compare our observations with the results reported in the literature