<![CDATA[Digital movie platforms expose users to thousands of titles, creating information overload and making personalized discovery increasingly important. This study aims to implement and evaluate a web-based movie recommender using a Neural Singular Value Decomposition (Neural SVD) model on the MovieLens Latest Small dataset. The research applies a quantitative computational experiment. After cleaning, 100,823 ratings from 610 users and 9,716 movies were encoded and divided into training and testing sets while ensuring that every movie appeared in the training data. Neural SVD was operationalized as embedding-based matrix factorization with user and item biases and a dot-product interaction. The model was evaluated using RMSE, MAE, R², a constant baseline, statistical significance testing, black-box testing, and a user questionnaire. The model achieved RMSE 0.8768, MAE 0.6733, and R² 0.2712, outperforming the baseline RMSE 1.0276 and MAE 0.8176. The difference was statistically significant (p = 0.0015), all eight functional scenarios were valid, and user acceptance reached 87.62%. These results indicate that Neural SVD can provide accurate personalized movie recommendations and can be integrated effectively into a Flask–React web architecture]]>
Copyrights © 2026
