PERSONALIZED MOVIE RECOMMENDATION SYSTEM USING ML

Abstract

Streaming platforms and online movie catalogues now offer tens of thousands of titles, and users frequently spend more time searching for something to watch than actually watching it. A recommendation system helps by selecting, from this large catalogue, a small number of movies that match the preferences of each individual user. This paper presents a personalised movie recommendation system that uses machine learning to suggest movies based on a user's past ratings, the ratings of similar users, and the content characteristics of the movies themselves. The system is built on a public movie ratings dataset containing explicit ratings given by users on a five-point scale, together with movie metadata such as title, release year, genres, and user-assigned tags. Additional descriptive information, including plot keywords, cast, and director, is used to build content profiles. The data is cleaned to remove duplicate ratings and titles with inconsistent identifiers, and users and movies with very few ratings are handled separately so that they do not distort the learned patterns. Three families of recommendation techniques are implemented and compared. Content-based filtering represents each movie by a vector of genres, tags, and keywords weighted with term frequency and inverse document frequency, and recommends movies similar to those a user has rated highly. Collaborative filtering uses user-based and item-based nearest neighbour methods as well as matrix factorisation through singular value decomposition to learn latent preference factors from the rating matrix. A hybrid model combines the collaborative prediction with content similarity so that new movies and users with few ratings can still receive sensible recommendations.