Design of Advertisement Creative Optimization and Performance Enhancement System Based on Multimodal Deep Learning

In order to improve the quality and efficiency of advertisement creative generation, an advertisement creative optimization and performance improvement system based on multimodal deep learning is designed. The system fuses three types of data, namely image, text and user behavior, and utilizes ResNet50 and BERT-base co-coding to achieve cross-modal feature alignment through a multi-layer attention fusion network, and a three-layer Transformer structure for deep semantic modeling. The experimental setup includes 12,00 sets of multimodal samples, the training batch is 128, the total number of training rounds is 200, the optimizer is AdamW, the learning rate is set to 1e-4, and the loss function contains the three-modal reconstruction loss and the cross-modal consistency loss. The system evaluates the performance of the advertisement creative through the multidimensional aspects of click rate prediction, sentiment consistency and image attractiveness. The experimental results show that the advertisement creative generation using this multimodal deep learning system is better than the traditional model, with a click-through rate accuracy of 0.681, an F1 value of 0.791, a PSNR of 28.94, and a composite score of 0.729, which is a significant improvement over the traditional method, demonstrating the advantages of the system in improving the quality of advertisement creative generation and user experience.