Personalized video recommendation with attentive multi-view learning and long-term and short-term interest modeling

Abstract Personalized video recommendation plays a crucial role in alleviating information overload on large-scale video platforms. However, existing approaches largely rely on identification (ID)-based features, which are insufficient for capturing the rich semantic information embedded in video content and user behaviors. To address this limitation, we propose a novel recommendation framework that jointly learns user and video representations through attentive multi-view feature modeling. In the video encoder, we design an attentive multi-view learning mechanism to generate unified video representations by integrating heterogeneous semantic information, including video category, title, content description, and cast information. In the user encoder, the framework models both long-term and short-term user interests: long-term interests are captured via graph-enhanced collaborative representation learning on the user–item interaction graph, with multi-layer graph propagation encoding higher-order structural dependencies between users and videos; short-term interests are modeled through an attention-enhanced bidirectional long short-term memory (Bi-LSTM) module to reflect users’ recent viewing preferences dynamically. Extensive experiments on a large-scale real-world dataset demonstrate that the proposed model consistently outperforms state-of-the-art baselines across all evaluation metrics, validating the effectiveness of attentive multi-view representation learning for enhancing personalized video recommendation performance.