<![CDATA[The growth of online video catalogs creates information overload that makes it difficult for users to discover movies matching their preferences. This study develops and evaluates two movie recommender approaches: content-based filtering using TF-IDF and cosine similarity, and collaborative filtering using Singular Value Decomposition (SVD). The dataset includes 4,803 movies with production-credit records and 100,004 user-rating interactions. The research stages comprise data cleaning and integration, content representation, latent-factor modeling, a 75:25 train-test split, and evaluation using Precision and Root Mean Squared Error. The content-based experiment achieved a Precision of 0.70 based on genre overlap, while director- and actor-based recommendations obtained a precision of 1.00 because the ranking features were identical to the relevance criteria. The SVD model achieved an RMSE of 0.9007 and generated personalized recommendations from user-rating patterns. The study contributes an evaluation of two baselines with different objectives, an analysis of potential bias in precision measurement, and an emphasis on consistent movie identifiers across datasets. Content-based filtering is effective for retrieving movies with similar metadata, whereas SVD captures latent user preferences. However, the results of the two approaches must be interpreted using metrics appropriate to their respective objectives. ]]>
Copyrights © 2026
