Emotion Analysis and Communication Effect Optimization for Cross-Cultural Documentaries Based on Convolutional Neural Networks

In the context of increasingly frequent cross-cultural communication, documentaries, as an important medium for cultural output, are largely limited by the differences in audiences' understanding of emotional content in different cultural backgrounds. In order to solve the problem of insufficient accuracy of existing emotion recognition methods in multicultural environments, this paper proposes a multimodal emotion analysis model that integrates convolutional neural networks (CNNs) and cultural embedding mechanisms. This method comprehensively utilizes multimodal inputs such as subtitle text, voice audio, and video key frames, and introduces cultural labels for multi-task joint training, thereby improving the model's ability to perceive emotional differences in cultural contexts. Experimental results show that the proposed model performs well in multilingual emotion recognition tasks, with an overall F1 value of 88.6%, which is better than the existing mainstream models. At the same time, the communication feedback mechanism constructed based on the recognition results can effectively improve the resonance and communication efficiency of documentaries among audiences from different cultures. This study provides new ideas for the integration of emotional computing and cross-cultural communication, and also lays a foundation for the application of multimodal emotion modeling technology in the field of intelligent media.