HPNewsRec: Hybrid perception entity-centric personalized news recommendation

In the digital era, accurate news recommendation systems are essential for enhancing user engagement and effective information delivery. Existing approaches have achieved notable progress by modeling semantic content, incorporating sentiment signals, or leveraging knowledge graphs. However, these methods are often studied in isolation and do not fully capture the complementary relationships among semantic information, user–news interactions, sentiment dynamics, and knowledge-based entity associations. As a result, their ability to model complex user interests and evolving news contexts remains limited. To address this issue, we propose a hybrid perception framework for personalized news recommendation, named HPNewsRec. The proposed model introduces a Hybrid Perception Model (HPM), which jointly integrates semantic representations, sentiment-aware signals, and knowledge graph-based contextual information within a unified framework. This design enables the model to better capture fine-grained user preferences and uncovers latent connections between emerging news entities and users’ historical interests. Extensive experiments on widely used benchmark datasets demonstrate that HPNewsRec consistently outperforms several strong baseline models across multiple evaluation metrics. The results indicate that jointly modeling heterogeneous signals through hybrid perception can effectively improve both recommendation accuracy and robustness, highlighting its potential for real-world news recommendation scenarios.