Hybrid Recommender System for Movies

U sklopu ovog završnog rada stvoren je hibridni sustav preporuka za filmove. Zove se hibridni jer pri stvaranju preporuke kombinira metode suradničkog filtriranja, kao i filtriranja baziranog na sadržaju. Suradničko filtriranje je bazirano na prikupljanju i obradi velike količine podataka o korisničkim sklonostima i predviđanju koji entiteti će se svidjeti korisniku, ovisno o korisnikovoj sličnosti s ostalim korisnicima. Filtriranje po sadržaju bazirano je na opisu entiteta značajkama i korisničkom profilu. Sustav je baziran na skupu podataka veličine oko 5000 filmova. Zamisao rada krajnjeg sustava je sljedeća: korisnik će unijeti film koji mu se svidio, a sustav će na osnovu tog unosa vratiti preporuku filmova koji bi se trebali svidjeti korisniku. Sustav prvo organizira i čisti podatke o filmovima, potom gradi sustav odnosa između značajki filmova, zatim u preporuku uključuje podatke o ocjenama drugih korisnika, te na kraju kombinira rezultate i daje konačnu preporuku. Rezultati sustava su ocijenjeni koristeći subjektivne ocjene o ispunjenju korisnika.The goal of this thesis was to make a hybrid movie recommender system. Hybrid system combines the methods of content based and collaborative filtering. Collaborative filtering is based on collecting large amounts of data about user preferences and inferring current user movie preferences, depending on his similarity with other users. Content based filtering is based on breaking movies down on discrete, analyzable features and making recommendations based on the similarities of those features. The system is based on movies database TMDB, which contains about 5000 movies. The system works as follows: First, the system organizes and cleans movie data, then a user selects a movie that he likes, then, the system analyzes the feature of that movie and makes a content based prediction, then the system includes the data about other users and makes a collaborative based prediction, then it combines both results to give a final prediction, represented by a list of movies that the user should like. Results were evaluated using subjective scores of user satisfaction