A heterogeneous graph neural network-based movie recommendation algorithm with improved GCNII

With the rapid growth in the number of films, users face the challenge of information overload, making it difficult to efficiently find movies that align with their personal interests. Consequently, personalized recommendation techniques play a crucial role in the field of information filtering. This paper proposes a movie recommendation algorithm based on an improved heterogeneous graph neural network using GCNII, which integrates user-movie interaction data, user features, and movie attributes to uncover deep relationships between users and movies. This approach mitigates the impact of data sparsity and enhances recommendation accuracy and diversity. Initially, the MovieLens dataset is preprocessed to extract user and movie features, constructing a heterogeneous user-movie interaction graph. The improved GCNII model is then applied to capture high-order semantic relationships within the interaction graph, while an attention mechanism and PairNorm regularization are incorporated to optimize the information propagation process and enhance model stability. Experimental results demonstrate that the proposed model achieves a HitRate@10 of 72.08% and a HitRate@5 of 68.23%, with an optimal RMSE value of 0.825, representing an average reduction of 1.09%.