Collaborative filtering recommender system for movies

Cilj ovog rada bio je razviti sustav preporuka za filmove temeljen na kolaborativnom filtriranju gdje se korisnicima pružaju personalizirane preporuke filmova u odnosu na slične korisnike. Najprije je obrađena tematika prikupljanja, obrade i transformacije podataka te trening i optimizacija neuronske mreže. Također, za implementaciju sustava korištena je konvolucijska graf neuronska mreža LightGCN i Pearsonov koeficijent korelacije. Rezultati oba modela su agregirani korištenjem tehnike učenja ansambla te su temeljito analizirani s ciljem ocjene točnosti i učinkovitosti razvijenog sustava preporuka. Za kraj, dan je pregled mogućih unaprjeđenja sustava koji bi uključivali dodatne podatke o korisnicima odnosno dodatne modele za preporuku.The aim of this work was to develop a movie recommendation system based on collaborative filtering where users are provided with personalized movie recommendations in relation to similar users. First of all, the topic of data collection, processing and transformation, as well as neural network training and optimization, was covered. The implementation of the system involved the use of LightGCN, a graph convolutional neural network, and Pearson's correlation coefficient. The results of both models were aggregated using ensemble learning techniques and were thoroughly analyzed with the aim of evaluating the accuracy and efficiency of the developed recommendation system. Finally, an overview of possible improvements to the system was given, which would include additional data on users or additional models for recommendation