基于多模态思维链对齐的大语言模型情感分析方法
By employing a large language model sentiment analysis method with multimodal thought chain alignment, the problems of modal inconsistency and conflict in multimodal sentiment analysis are resolved, thereby improving the accuracy and interpretability of sentiment analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HEFEI UNIV OF TECH
- Filing Date
- 2026-01-23
- Publication Date
- 2026-07-17
AI Technical Summary
Existing multimodal sentiment analysis methods lack explicit reasoning mechanisms, making it difficult to explain the basis of model decisions. Furthermore, inconsistencies or conflicts exist in the reasoning information between different modalities, resulting in insufficient accuracy and interpretability of sentiment analysis.
We employ a large language model sentiment analysis method with multimodal thought chain alignment. By acquiring and preprocessing text, visual, and speech modal data, we utilize a large language model for modality-specific thought chain reasoning, employ an attention weighting mechanism for fusion, and construct a sentiment alignment loss function to resolve modality conflicts.
It improves the accuracy, robustness, and interpretability of multimodal sentiment analysis, and can effectively align and coordinate different modalities when they conflict, thereby enhancing the accuracy and stability of sentiment analysis.
Smart Images
Figure CN121980314B_ABST