MGCAM: Sentiment analysis model of e-commerce reviews based on multi-granularity and cross-attention mechanism

With the continuous development of e-commerce platforms, more and more consumers make online shopping, and product reviews contain consumers' potential shopping experience and satisfaction. E-commerce platforms can improve consumers' satisfaction by accurately identifying consumers' emotional tendencies, thus contributing to the sustainable development of e-commerce platforms. To accurately identify the potential emotional tendency of consumers in e-commerce reviews, this paper proposes a Multi- Granularity and Cross- Attention Mechanism based e-commerce review emotion classification model (MGCAM). Firstly, a multi-granularity feature extraction module based on CNN and LSTM is proposed in this model, and the potential features in comments are mined from two levels of fine granularity and coarse-granularity respectively. Secondly, in order to further enrich the features, the self-attention mechanism model is used to extract text comments. Finally, in order to improve the robustness of the model, a multi-head cross-attention mechanism is used to cross-fuse the output representations of the above two parallel modules, and the final output representations are used for emotion classification. Through an experimental analysis of e-commerce platform reviews for cold chain products, the proposed MGCAM model achieved an F1 score of 91.79%, surpassing all comparative algorithms and confirming its effectiveness.