Fusion with GCN and SE-ResNeXt Network for Aspect Based Multimodal Sentiment Analysis
Jun Zhao, Fuping Yang · 2023
Aspect Based Multimodal Sentiment Analysis to determine the sentiment polarity of each aspect mentioned in multimodal posts or comments. However, many models fail to explain the related syntactic constraints and long-term word dependencies, as well as lack image feature extraction. In view of the above problems, we propose a Fusion with GCN and SE-ResNeXt Network(FGSN), which constructs a graph convolution network on the dependency tree of sentences to obtain the context representation and aspects words representation by using syntactic information and word dependency. Also, we use the pretrained network SE-ResNeXt101 to extract image features and generate visual representations. Moreover, the position attention and channel attention mechanism are used to obtain the image features. Finally, we fuse image features and text features to classify sentiment polarity. The experimental results show that our model performs well on the two datasets of TWITTER-2015 and TWITTER-2017.