<p>Recent research has demonstrated the efficacy of leveraging multi-grained user-news matching signals, such as word and news extracted via pretrained language models, to enhance news recommendation systems and mitigate the cold-start issue. However, existing approaches primarily focus on text-based recommendation models, treating user histories and candidate texts as sole inputs. Despite the potential benefits of incorporating diverse side information to enhance recommendation performance, and the abundance of such information in real-world news recommendation scenarios, current methodologies often overlook the integration of additional contextual cues into pretrained language models. In this study, we address this gap by investigating the integration of news category information into Transformer encoders for news recommendation, without introducing new parameters. We propose a novel architecture called Category-Enhanced Local- and Global-Attention Transformer Encoder (CETen), which aims to better capture user-news matching signals. Unlike previous methods, CETen incorporates a category-level attention module to facilitate local word attention within news articles of the same category. Furthermore, it introduces category-based special token prompts within the self-attention mechanism of the Transformer to enhance the modeling of text semantics. Experimental results on the MIND-small and MIND-large news datasets substantiate the superiority of our proposed model over existing state-of-the-art methods, underscoring its effectiveness in news recommendation tasks.</p>