Recommendation systems are vital for mitigating information overload. Knowledge graphs enhance these systems, but traditional models overlook dynamic user behavior and context. This paper proposes a model integrating contextual information and attention to address these issues. The proposed model utilizes users’ historical data and incorporates various contextual factors, such as film attributes and metadata, through an attention mechanism. This mechanism dynamically adjusts the weight of context information and employs a multi-layer neural network to fit features and relationships accurately. As a result, the model better captures users’ interests and preferences, improving recommendation accuracy and user experience. Experimental results, with a recommended length of 10, yield Recall indices of 0.1934, 0.1379, and 0.0801 for MovieLens-1M, Amazon-book, and Last.FM datasets, respectively. Optimal performance is achieved at an embedding dimension of 32.