Hybrid Recommender System For Movies

Diplomski rad opisuje model preporučiteljskog sustava zasnovanog na sadržajnim značajkama i model preporučiteljskog sustava zasnovanog na suradničkom filtriranju. U modelu zasnovanom na sadržajnim značajkama definiraju se profil proizvoda, profil korisnika i funkcija sličnosti. U modelu zasnovanom na suradničkom filtriranju definiraju se susjedstvo, funkcija sličnosti i izračun predviđanja. Uvodi se hibridni model za preporuku filmova. Osnova modela je model zasnovan na sadržajnim značajkama, a unaprjeđuje se modelom zasnovanom na suradničkom filtriranju. Hibridni model umanjuje nedostatke: „hladni start“ (engl. Cold start) i pre-specijaliziranost, te koristi prednosti za bolju preporuku filmova. Pročišćava se i prilagođava skup podataka MovieLens. MovieLens sadrži 1,000,209 korisničkih ocjena filmova, te podatke o 3,900 filmova. Programski se ostvaraje model za preporuku filmova zasnovan na sadržajnim značajkama. Računa se točnost modela. Programski se ostvaruje hibridni model za preporuku filmova. Računa se točnost modela i uspoređuje se sa točnosti modela zasnovanog na sadržajnim značajkama. Komentiraju se rezultati.The final thesis presents a comprehensive exploration of two recommender system models: a content-based model and a collaborative filtering model. The content-based model encompasses the definition of product and user profiles, as well as the establishment of a similarity function. The collaborative filtering model introduces the concept of neighborhood, similarity function, and prediction calculation. To enhance the recommendation process for movies, a hybrid model is introduced, building upon the foundation of the content-based model and incorporating the strengths of collaborative filtering. This hybrid approach effectively addresses challenges such as the cold start problem and over-specialization, resulting in improved movie recommendations. The MovieLens dataset is carefully refined and adapted to ensure its suitability for the study. This dataset encompasses a wealth of information, including 1,000,209 user ratings for movies and data pertaining to 3,900 films. Through the implementation of a programmatic movie recommendation model based on content features, the accuracy of the model is evaluated. Additionally, a hybrid model for movie recommendation is implemented, and its accuracy is compared to that of the content-based model. Finally, the results are thoroughly examined and discussed to provide valuable insights and conclusions