With the rapid development of new media and the rise of self media, the copyright of content belongs to users, and ordinary users can also create corresponding content. However, the creation of popular science works is a very scientific and rigorous work that cannot be completed by ordinary users. The output of high-quality popular science content still relies on professional editors from popular science media organizations or relevant research institutions, as well as the scientific and technological departments of authors of popular science publications. Therefore, how to create timely and professional popular science content has become a new topic. This article explored the creation and recommendation of new media content based on artificial intelligence. The Naive Bayes algorithm was used to classify new media content, and mixed similarity was used to calculate the similarity between users and content, achieving the creation and recommendation of new media content. The experimental results showed that the recommendation accuracy of the hybrid recommendation algorithm was higher than that of the collaborative filtering algorithm, and the difference in accuracy between the two was at least 2%.