Fusing Fine-Grained Information of Sequential News for Personalized News Recommendation

In this paper, we propose a novel method that fuses Fine-grained Information of Sequential News for personalized news recommendation (FISN). FISN comprises three primary modules: news encoder, clicked news optimizer and user encoder. The news encoder uses fine-grained information to learn accurate news representations. The clicked news optimizer introduces multi-headed self-attention and positional encoding techniques to optimize the clicked news representation. The user encoder uses news-level attention to learn user representations. Extensive experimental results demonstrate that FISN outperforms many baseline approaches in terms of metrics for real datasets.