辽宁大学学报（自然科学版）2026，Vol.53Issue(1)：51-61,11.
文本导向的多任务多模态情感感知分析模型
Text-Guided Multi-task Multimodal Emotion Perception Analysis Model
摘要
Abstract
To address the limitations of existing multimodal sentiment analysis models in underemphasizing the text modality and utilizing context,a multi-task multimodal emotion perception analysis model that integraces text augmentation and cross-modal interactive attention mechanisms is proposed.Specifically,we introduce a GPT-based text augmentation technique to enhance the recognition of emotional words and reinforce the text modality.A cross-modal interactive attention mechanism is then employed to fuse visual,audio and textual information effectively.Moreover,a homoscedastic uncertainty loss function is applied to optimize weight adjustments across multiple tasks.On the CMU-MOSI and CMU-MOSEI datasets,the proposed model achieves ACC2 and F1 scores of 87.2%and 85.8%respectively,outperforming the baseline model.This demonstrates the effectiveness of the proposed model and its significant improvement in multimodel perception analysis performance.关键词
多模态情感分析/情感词感知/文本信息增强/多任务学习Key words
multimodal sentiment analysis/emotional word perception/text information enhancement/multi-task learning分类
信息技术与安全科学引用本文复制引用
臧洁,李翔,卢睿,廖慧之,任赛赛,卢珊..文本导向的多任务多模态情感感知分析模型[J].辽宁大学学报（自然科学版）,2026,53(1):51-61,11.基金项目
辽宁省科技厅应用基础研究(2023JH2/101300134) （2023JH2/101300134）
2025辽宁大学研究生优质课程建设与教改研究项目(YJG202501075) （YJG202501075）
