MGMFN: Multi-graph and MLP-mixer fusion network for Chinese social network sentiment classification

Sentiment analysis (SA) in social networks plays a vital role in the current information industry era. It has a broad application basis and crucial practical significance for research in public opinion analysis, hot spot mining, and product recommendation. The current research on SA has gradually expanded from convolutional neural networks to graph neural networks since the graph structure can contain more text sentiment features. However, the improvement is limited due to the informal expressions and complexity of online reviews, and lack of fusing the context information of the statement, the dependency parsing results, and the syntactic results. To overcome these challenges, we aim to study the application of graph neural network in SA, investigate the sentiment features of the text from multiple views for the sentiment classification of Chinese text and propose a multi-graph and MLP-Mixer fusion network (MGMFN) model. This model effectively integrates the contextual information, syntactic information, and semantic information of Chinese text and considers the complementarity among contextual information, syntactic structures, and semantic correlations. The mechanism of the MLP-Mixer is utilized to enhance long-range semantic dependencies in the text and strengthen the spatially representational power of multi-head attention. In the experiment, the Macro_F_measure and Micro_F_measure of our model reach 83.72