Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Янв. 7, 2025

In the rapidly evolving field of personalized news recommendation, capturing and effectively utilizing user interests remains a significant challenge due to vast diversity dynamic nature interactions with content. Existing recommendation models often fail fully integrate candidate items into interest modeling, which can result in suboptimal accuracy relevance. This limitation stems from their insufficient ability jointly consider history characteristics modeling process. To address this challenges, we propose Multi-view Knowledge Representation Learning (MKRL) framework for leverages multi-view encoder candidate-aware attention mechanisms enhance modeling. Unlike traditional methods, MKRL incorporates articles directly process, enabling model better understand predict preferences based on both historical behavior potential new is achieved through sophisticated architecture that blends mechanisms, together capture more holistic view interests. The innovatively integrates convolutional neural networks multi-head modules intricate contextual information news, allowing recognize fine-grained patterns. dynamically weighs relevance, enhancing accuracy. Additionally, MKRL's approach represents different perspectives, richer recommendations. Extensive experiments three real-world datasets demonstrate our proposed outperforms state-of-the-art baselines accuracy, validating its effectiveness.

Язык: Английский