Text-based Multimodal Sentiment Analysis: Missing Modality Prompt Learning and Hyper-modality Representation

Sentiment analysis is a critical task in natural language processing, aimed at identifying emotional tendencies from text. With the advent of multimodal technologies, sentiment analysis has expanded from a unimodal text-based approach to multimodal sentiment analysis, which integrates information from multiple modalities such as text, vision, and audio to enhance the accuracy of sentiment recognition. However, in practical applications, multimodal data often suffer from missing modalities, presenting significant challenges to sentiment analysis.In this paper, we propose a text-based multimodal sentiment analysis method that focuses on addressing the issues of missing modality prompt learning and hyper-modal representation. The proposed method introduces a missing modality prompt mechanism, which supplements the missing features of non-text modalities through generated prompts, thereby enhancing the model's robustness to incomplete data. Additionally, by incorporating a hyper-modal learning module, it learns conflict-suppression representations from visual and audio features under the guidance of text features. This study provides new insights into multimodal sentiment analysis, especially in handling missing modalities and constructing hyper-modal representations, offering valuable references for future research.