Personalized recommender systems represent intelligent algorithms and decision-making mechanisms that are meticulously crafted on the bedrock of extensive datasets. In the realm of online recommender systems, the process involves tailoring recommendations to individual users by harnessing their historical behavioral data, which encompasses activities such as browsing, watching, purchasing, and rating. Simultaneously, these systems delve into the content information associated with the items on offer, including introductory details, specific features, and applicable scenarios. The essence of personalized recommendations lies in the continuous refinement of the system's output based on user interaction feedback. In recommender systems, understanding the relationships between items is crucial. We introduce a sophisticated multimodal cooperative learning strategy that explicitly captures the interconnections among diverse modalities, elevating the representation quality of each modality.