This paper designed and constructed a multimodal teaching content recommendation system based on artificial intelligence technology, which was mainly used to solve the problems of personalization deficiency, single mode dependency and context neglect in traditional teaching resource recommendation systems. In the first recommendation performance evaluation experiment, the recommendation performance of the Transformer algorithm was tested, and the experiment found that the Transformer achieved accuracy, recall, and F1 score of 85%, 80%, and 82%, respectively. The experimental results showed that the fusion strategy based on Transformer outperformed early and late fusion strategies in recommendation accuracy of 80%, recall of 75%, and F1 score of 77%. The above experimental results showed that the Transformer based multimodal teaching content recommendation system could effectively improve the accuracy and user satisfaction of recommendations, and also demonstrated significant advantages in processing complex multimodal data and adapting to diverse teaching fields.