Visual Sentiment Analysis for Review Images with Item-Oriented and User-Oriented CNN by Introducing CBAM
Ying Liu, Zhe Wang, Jie Fang · 2022 4th International Conference on Natural Language Processing (ICNLP) · 2022
Sentiment analysis is a common scenario in natural language processing, which plays a key role in guiding product updates and iterations. Recently, with the diversification of the web, images have gradually become an important data form in the social media domain. Most of the previous visual sentiment analysis of social media have only considered the influence of image factors but ignored item factors and user factors. In this case, useful sentiment feature information can not be deeply explored, and which would limit the effectiveness and stability of the system. In this paper, we propose a iCVS-CNN based on item factors and a uCVS-CNN model based on user factors for the visual analysis of social media comment images. The models focus on learning feature representations in different subspaces by sequentially inferring the attention graph along two independent dimensions (channel and space) through a convolutional block attention module (CBAM). It is shown experimentally that the CBAM module can help to focus more on the sentiment features, which verifies the effectiveness of the model.