Research on Cross-Cultural Communication Effect Modeling and Optimization Method Based on Multimodal Corpus and Multi-Objective Reinforcement Learning

To address the challenge of insufficient communication adaptability of Chinese brand stories in multicultural contexts, this paper proposes a cross-cultural communication effect modeling and optimization method based on multimodal deep representation and multi-objective reinforcement learning. The proposed framework leverages textual, visual, and auditory modalities using multilingual BERT, ResNet-152, and VGGish respectively. A cross-modal attention mechanism is applied for semantic alignment, and a Cultural Semantic Embedding (CSE) module is introduced to quantify cultural adaptation. For propagation strategy optimization, a Proximal Policy Optimization (PPO)-based multi-objective reinforcement learning algorithm is designed, with cognitive acceptance, affective resonance, and cultural misreading rate as joint reward functions. A large-scale multimodal corpus covering 30 Chinese international brands and five cultural regions is constructed. Experimental results show that the proposed model outperforms baseline models in all communication metrics, achieving a 4.5 user satisfaction score and reducing bounce rate by 9.4%. Compared to traditional models, the framework demonstrates superior adaptability, interpretability, and scalability. This study not only advances the application of deep multimodal modeling and reinforcement learning in intelligent communication systems, but also contributes to the interdisciplinary integration of engineering computation and human-centered cultural communication.