基于主模态增强与多阶段融合的鲁棒多模态情感分析方法
By constructing a robust multimodal sentiment analysis method that combines main modality enhancement and multi-stage fusion, the performance degradation problem of multimodal sentiment analysis under modality loss and noise interference is solved, and stable sentiment recognition and feature discrimination are achieved in complex scenarios.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- EAST CHINA JIAOTONG UNIVERSITY
- Filing Date
- 2026-05-12
- Publication Date
- 2026-07-17
AI Technical Summary
Existing multimodal sentiment analysis methods suffer from performance degradation in scenarios with missing modalities and noise interference. They lack systematic missing modality compensation and noise suppression mechanisms, have insufficient dynamic adaptation capabilities of the main modality, and their multimodal fusion is not deep enough, making it difficult to maintain stable sentiment recognition performance across all scenarios.
We construct a robust multimodal sentiment analysis method based on master modality enhancement and multi-stage fusion, including an embedded encoder, an integrity-checking encoder, a text-guided cross-modal collaborative enhancer, a hierarchical conditional cue generator, a weighted fusion generator, and a modality-specific reconstructor. Through multi-stage processing, we improve the robustness and adaptability of the model.
It effectively solves the feature processing problem when the main modality is missing, realizes lightweight semantic compensation and feature representation quality assurance, improves the adaptability and robustness of the model in complex scenarios, and significantly enhances the discriminative ability of cross-modal features.
Smart Images
Figure CN122173655B_ABST