End-to-end recommendation engine with content-based filtering (TF-IDF), collaborative filtering (SVD), and popularity-based ranking on MovieLens 100K.
Built an end-to-end movie recommendation system using the MovieLens 100K dataset, implementing content-based filtering (TF-IDF), collaborative filtering (SVD), and popularity-based ranking to compare different recommendation strategies. Built an interactive Streamlit application for exploring personalised recommendations and evaluated model performance using RMSE, Precision@10, NDCG, and genre consistency.
Encodes movie titles and genres into TF-IDF vectors and recommends similar items using cosine similarity.
Matrix factorisation via Surprise library. Learns latent user-item interaction patterns from the ratings matrix.
Baseline ranking by average rating and vote count. Useful for cold-start and non-personalised recommendations.
Rating prediction accuracy for SVD model
How many of the top-10 recommendations are relevant
Ranking quality — are the best items ranked highest?
Do recommended movies share genres with the input?