Emotion Classification Representation Method Based on Multimodal Hierarchical Feature Mapping and Fusion

With the development of mobile internet and 5G technology, social media platforms have rapidly evolved, enabling users to share opinions and express demands online. In this context, single-modal information representation is insufficient to fully capture the richness of internet content. Therefore, this paper proposes a multi-modal hierarchical feature mapping and fusion representation method (MHFMFR). This method combines the text model RoBERTa and the image model DenseNet, effectively mapping and fusing multi-modal data through a feature reuse strategy, thus improving the accuracy of sentiment classification. Experiments demonstrate that this method outperforms existing baseline models on the Weibo sentiment and MVSA-Multiple datasets.