Target-Oriented Multimodal Sentiment Classification Based on Multi-Head Attention and Graph Neural Network

Abstract

With the rapid development of the Internet, multimodal information on various social media platforms has proliferated. In this paper, a target-oriented multimodal sentiment classification model based on the multi-head attention and graph neural network was proposed. Specifically, the model constructed a two-channel semantic aggregation network for texts, connected it with the syntactic aggregation network, and then obtained the text features using the attention mechanism. For images, the object features and scene features were extracted respectively by the graph neural network. Finally, the multi-modal deep fusion was realized through the multi-head attention mechanism, and the result was used for target-oriented multimodal sentiment classification. Through repeated experiments on two public datasets, the effectiveness of the proposed model was verified.