Abstarct: With the rapid expansion of social media, platforms such as Twitter have become influential environments for the emergence and diffusion of news streams. These streams typically evolve around specific topics or hashtags and consist of continuous user interactions that attract public attention over successive time intervals. Effectively guiding such streams is important not only for increasing public awareness but also for applications such as digital advertising, crisis communication, and public opinion formation. Although extensive research has been conducted on information diffusion and recommender systems, an integrated frxamework that simultaneously combines target user recommendation and automated content generation for the purpose of orchestrating news streams has not yet been sufficiently explored. This research proposes a novel frxamework for generating news streams on Twitter by integrating a hybrid recommender system with large language models. The proposed frxamework employs collaborative filtering and content-baxsed filtering to recommend relevant content and identify target users with high potential for engagement and influence. By analyzing users’ behavioral feedback and evaluating the semantic relevance between user-generated content and the campaign topic, the system generates context-aware messages aligned with the selected theme. In the initial stage, the frxamework leverages the coordinated activity of bot-controlled accounts within a controlled network to create early awareness around the target topic. As interactions increase and real users begin to participate, the diffusion process expands and the news stream gradually evolves through network-level engagement dynamics. To evaluate the effectiveness of the proposed approach, two key metrics—hashtag frequency and user participation level—were used, and the performance of the frxamework was compared with a baxseline method. Experimental results show that the total number of actions performed by bot-driven accounts increased from 7,112 in the baxseline method to 15,074 using the proposed frxamework. Furthermore, real user participation increased substantially, from 2,214 posts generated by 2,088 users to 134,665 posts generated by 13,490 users. In addition, the generated hashtag achieved significant visibility, appearing among the top 20 trending topics on Twitter with more than 134,000 posts. These results demonstrate that integrating recommender systems with large language models can effectively enhance user engagement, improve information diffusion, and facilitate the formation and expansion of news streams in social networks