A Construction Method for Personalized Movie Recommendation System Based on Embedding
Wenwei Zeng · Applied and Computational Engineering · 2025
Personalized recommendation systems are very important. They can improve user experience and help platforms make more money. This study introduces a movie recommendation model based on deep learning and embedding techniques. We train this model using the MovieLens-25M dataset. Different from traditional collaborative filtering methods, our system changes users and movies into a low-dimensional latent space. It uses a multi-layer perceptron to catch complex nonlinear interactions. To solve the problems of data sparsity and cold-start, we use negative sampling and leave-one-out evaluation. The experimental results show that our model works well. The Hit Ratio @10 is 0.85. This proves that embeddings are useful for learning user preferences. Although our model gives good recommendations, it still has limitations. It does not use multimodal data and does not model time changes. In future work, we will use Transformer models, large language models, and multimodal data. These methods can make recommendations more personalized, easier to understand, and more adaptable in dynamic environments.