Research on Cross-language Convergence Media Text Sentiment Analysis Technology

This study aims to advance cross-language sentiment analysis for social media texts, addressing the challenge of limited annotated corpora in low-resource languages, which hinders effective sentiment classification critical for understanding public opinion across diverse regions. We propose a novel framework integrating a reinforcement learning module with generative adversarial cross-language knowledge transfer to minimize feature distribution gaps, and a bilinear attention-based paired-language interaction module, inspired by multimodal learning, to capture emotional semantic similarities. Experimental results on English-French, English-German, and English-Hindi Twitter datasets show significant improvements, with the ADDS-MBAN model achieving up to 76.2% accuracy and 70.4% F1 score in English-French tasks, outperforming baseline models like Multi-BERT and MT-DAN. These advancements enhance sentiment classification in low-resource settings, enabling robust cross-lingual public opinion analysis.