Deep Learning-Based Feature Selection for Movie Recommendation Services

Accurate movie recommendation systems rely heavily on selecting informative user and item features to capture underlying preferences effectively. This study proposes a deep learning-based feature selection framework integrated into a recommendation model, designed to automatically identify and prioritize the most relevant attributes from the MovieLens 1M dataset. The model filters out noisy and redundant features by leveraging attention mechanisms and differentiable selection layers, thereby improving prediction accuracy. Experimental results demonstrate that the proposed approach outperforms classical matrix factorization and neural collaborative filtering baselines regarding RMSE and MAE metrics. Additionally, the learned feature importance scores are consistent, interpretable, and align with domain knowledge, highlighting age, genre, and occupation factors. These findings underscore the value of incorporating deep feature selection techniques in recommendation systems to enhance performance and transparency.