In short, the French product reviews on the cross-border e-commerce platform are full of users’ emotions and feelings. Due to the complex grammar and implicit expressions in French, the old ways of sentiment analysis have been less effective. Therefore, this paper builds an integrated sentiment analysis and satisfaction prediction model named FSA-Net for French reviews. Through French text preprocessing, bidirectional GRU sequence modelling and a multi-scale emotion perception mechanism, the joint recognition of explicit and implicit emotions has been achieved. At the same time, a satisfaction fusion layer and a gating adjustment strategy are introduced to improve the non-linear mapping ability of emotions to scores. The two sets used for evaluation were the French-Amazon dataset (about 12,000 entries) and the self-built FRC dataset (8,200 entries). FSA-Net had an accuracy of 0.889 and an AUC of 0.944 for FAR, and an AUC of 0.931 for FRC. All of them outperformed the Bi-LSTM baseline model; the mean training deviation was still 0.0387, but stability had increased significantly. Based on the above results, the proposed model can improve the accuracy of emotion recognition and satisfaction prediction for French reviews effectively, and thus provides a practical technical path for the localisation intelligent analysis of cross-border e-commerce platforms.
