Efficient Multimodal Sentiment Analysis Model with Mamba and MobileNetV4

Due to the quick growth of social networks, users increasingly express sentiment through a combination of images and text on online platforms. Sentiment analysis of these data resources has significant implications across various fields. Multimodal sentiment analysis, which integrates multiple data modalities, can effectively overcome the limitations of unimodal approaches. This paper proposes a multimodal sentiment analysis model that leverages Mamba for text feature extraction and MobileNetV4 for image feature extraction, using a twodimensional tensor fusion network to capture deep interactions between modalities. The model effectively preserves and extracts complementary information from both modalities, enhancing sentiment classification's robustness and accuracy. Experimental results on the public MVSA dataset show that the proposed model outperforms others in sentiment analysis tasks.