Papers with vision-language models pre-
Vision-Language Pre-Training for Multimodal Aspect-Based Sentiment Analysis (2022.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to multimodal Aspect-Based Sentiment Analysis (MABSA) ignore crossmodalalignment and use pre-trained visual and textual models. |
| Approach: | They propose a multimodal multimodal encoder-decoder framework for MABSA that uses a unified multimodal decoder architecture for all the pretrainingand downstream tasks. |
| Outcome: | The proposed framework outperforms state-of-the-art approaches on three MABSA subtasks. |