Learning fine-grained representation with token-level alignment for multimodal sentiment analysis

Xiang Li, Haijun Zhang, Zhiqiang Dong, Xianfu Cheng, Yun Liu, Xiaoming Zhang

Abstract: Highlights•Propose fine-grained multimodal fusion network for sentiment analysis.•Extract fine-grained sentiment representations using fewer denoising tokens.•Perform token-level alignment to facilitate representation learning and fusion.•Generate consistent multimodal representations via correlation-aware fusion.