To address the problem of insufficient modal alignment before feature fusion in multimodal sentiment analysis, an emotion score difference-guided hybrid supervised contrastive learning method that operates on both intra-modal and cross-modal relationships is proposed. Specifically, the method uses differences between sample labels to construct positive and negative sample pairs, where samples with smaller differences are regarded as positive pairs and those with larger differences are regarded as negative pairs. Based on this, supervised contrastive learning is performed to capture affective semantic relationships within and across modalities, thereby encouraging features from different modalities to align more consistently in the semantic space. In addition, a dynamic attention-weighted fusion mechanism is introduced to optimize feature fusion by adaptively estimating the contribution of each modality for the current sample. Comparative experiments on different datasets demonstrate the effectiveness of the proposed method. Compared with baseline models, the proposed method achieves a maximum improvement of 19.98% in five-class accuracy and 0.167 in Pearson correlation on the CMU-MOSI dataset; a maximum improvement of 3.07% in seven-class accuracy on the CMU-MOSEI dataset; and maximum improvements of 7.66% and 9.19% in binary and three-class accuracy, respectively, as well as 7.71 in F1 score, on the CH-SIMS dataset.
GandhiA, AdhvaryuK, PoriaS, et al. Multimodal sentiment analysis: A systematic review of history, datasets, multimodal fusion methods, applications, challenges and future directions[J]. Information Fusion, 2023, 91: 424-444.
[2]
NguyenT H, ShiraiK. Topic modeling based sentiment analysis on social media for stock market prediction[C]//Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). Beijing: Association for Computational Linguistics, 2015: 1354-1364.
[3]
ZengY, MaiS J, HuH, et al. Multimodal sentiment analysis in the wild: A survey[J]. IEEE Transactions on Neural Networks and Learning Systems, 2022, 33(10): 4795-4814.
[4]
HuanRuohong, ZhongGuowei, ChenPeng, et al. UniMF: A unified multimodal framework for multimodal sentiment analysis in missing modalities and unaligned multimodal sequences[J]. IEEE Transactions on Multimedia, 2024, 26: 5753-5768.
[5]
PanY, ShenY, LiangP P, et al. Cross‑modal conflict resolution for multimodal sentiment analysis[C]//Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Singapore: ACL, 2023: 1895‑1906.
[6]
ChenBingzhi, CaoQi, HouMixiao, et al. Multimodal emotion recognition with temporal and semantic consistency[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2021, 29: 3592-3603.
[7]
ZadehA, ChenMinghai, PoriaS, et al. Tensor fusion network for multimodal sentiment analysis[C]//Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing. Copenhagen: Association for Computational Linguistics, 2017: 1103-1114.
[8]
HazarikaD, ZimmermannR, PoriaS. MISA: Modality-invariant and-specific representations for multimodal sentiment analysis[C]//Proceedings of the 28th ACM International Conference on Multimedia. Seattle: ACM, 2020: 1122-1131.
[9]
TsaiY H, BaiShaojie, LiangP P, et al. Multimodal transformer for unaligned multimodal language sequences[C]//Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. Florence: Association for Computational Linguistics, 2019: 6558-6569.
[10]
SunLicai, LianZheng, LiuBin, et al. Efficient multimodal transformer with dual-level feature restoration for robust multimodal sentiment analysis[J]. IEEE Transactions on Affective Computing, 2024, 15(1): 309-325.
[11]
WuDaiqing, YangDongbao, ShenHuawen, et al. Resolving sentiment discrepancy for multimodal sentiment detection via semantics completion and decomposition[J]. Pattern Recognition, 2026, 172: 112719.
[12]
HePing, QiHuaying, WangShiyi, et al. Cross-modal sentiment analysis of text and video based on Bi-GRU cyclic network and correlation enhancement[J]. Applied Sciences, 2023, 13(13): 7489.
[13]
YangJianing, WangYongxin, YiRuitao, et al. MTAG: Modal-temporal attention graph for unaligned human multimodal language sequences[C]//Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Virtual Event: Association for Computational Linguistics, 2021: 1009-1021.
[14]
SijieMai, XingSonglong, HeJiaxuan, et al. Multimodal graph for unaligned multimodal sequence analysis via graph convolution and graph pooling[J]. ACM Transactions on Multimedia Computing, Communications, and Applications, 2023, 19(2): 1-24.
[15]
Sharma Vishal、Rishu Null、Kukreja Vinay,et al. A novel integration of simple framework for contrastive learning of visual representations and data-efficient image transformer for diabetic macular edema diagnosis[J]. Iran Journal of Computer Science, 2025, 8(4): 1615-1645.
[16]
RadfordA, KimJ W, HallacyC, et al. Learning transferable visual models from natural language supervision[C]//Proceedings of the 38th International Conference on Machine Learning. Cambridge:PMLR, 2021: 8748-8763.
[17]
LiJunnan, SelvarajuR, GotmareA, et al. Advances in neural information processing systems[C]//Advances in Neural Information Processing Systems. New York: Curran Associates, Inc., 2021, 34: 9694-9705.
[18]
ZadehA, ZellersR, PincusE, et al. Multimodal sentiment intensity analysis in videos: Facial gestures and verbal messages[J]. IEEE Intelligent Systems, 2016, 31(6): 82-88.
[19]
Bagher ZadehA, LiangP P, PoriaS, et al. Multimodal language analysis in the wild: CMU-MOSEI dataset and interpretable dynamic fusion graph[C]//Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Melbourne: Association for Computational Linguistics, 2018: 2236-2246.
[20]
YuWenmeng, XuHua, MengFanyang, et al. CH-SIMS: A Chinese multimodal sentiment analysis dataset with fine-grained annotation of modality[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Virtual Event: Association for Computational Linguistics, 2020: 3718-3727.
[21]
LiuZhun, ShenYing, LakshminarasimhanV B, et al. Efficient low-rank multimodal fusion with modality-specific factors[C]//Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). Melbourne: Association for Computational Linguistics, 2018: 2247-2256.
[22]
RahmanW, HasanM K, LeeS, et al. Integrating multimodal information in large pretrained transformers[C]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Virtual Event: Association for Computational Linguistics, 2020: 2359-2369.
[23]
HanWei, ChenHui, PoriaS. Improving multimodal fusion with hierarchical mutual information maximization for multimodal sentiment analysis[C]//Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing. Virtual Event: Association for Computational Linguistics, 2021: 9180-9192.
[24]
ZhangHaoyu, WangYu, YinGuanghao, et al. Learning language-guided adaptive hyper-modality representation for multimodal sentiment analysis[C]//Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Singapore:Association for Computational Linguistics, 2023: 756-767.
[25]
WangDi, GuoXutong, TianYumin, et al. TETFN: A text enhanced transformer fusion network for multimodal sentiment analysis[J]. Pattern Recognition, 2023, 136: 109259.
[26]
YuWenmeng, XuHua, YuanZiqi, et al. Learning modality-specific representations with self-supervised multi-task learning for multimodal sentiment analysis[C]//Proceedings of the AAAI Conference on Artificial Intelligence. Washington, DC: AAAI Press, 2021: 10790-10797.
[27]
MaFeipeng, ZhangYueyi, SunXiaoyan. Multimodal sentiment analysis with preferential fusion and distance-aware contrastive learning[C]//2023 IEEE International Conference on Multimedia and Expo (ICME). Brisbane: IEEE, 2023: 1367-1372.