Adaptive Feature Fusion for Deep and Shallow Movie Recommendation

With the rapid development of Internet technology and smart devices, personalized movie recommendation systems have played an increasingly important role in enhancing user viewing experiences. To address issues such as cold-start, data sparsity, and limited non-linear feature extraction in traditional recommendation methods, this paper proposes a hybrid recommendation model based on the fusion of shallow networks and deep autoencoders (Shallow&AutoEncoder). The model captures explicit interaction features between users and items through a shallow network while learning latent nonlinear implicit features using a deep autoencoder. An adaptive fusion mechanism is designed to dynamically adjust the importance of shallow and deep features, thereby improving recommendation performance. Extensive experiments conducted on the MovieLens- 100 K dataset demonstrate that the proposed method outperforms traditional collaborative filtering, SVD, SVD++, and single autoencoder models in terms of MAE and RMSE metrics. Furthermore, ablation studies validate the contribution of each module to the overall performance improvement. This research provides an effective new approach for personalized recommendation systems and shows promising application potential.