Research on Comment Data Mining of Cross-Border E-Commerce of Belt and Road Countries Based on Cross-Lingual Sentiment Classification Technology

A sentiment recognition model of evaluation aspects based on improved attention mechanism and multi-task learning is proposed to mine the e-commerce review data in cross-border trade more deeply. The emotion recognition model based on Bi-LSTM is taken as the basic emotion recognition model, and the emotion recognition level of the model is further improved by adding attention mechanism to the emotion recognition model and constructing multi-task learning model. The experimental results show that compared with other emotion recognition models, the designed sentiment recognition model based on improved attention mechanism and multi-task learning has higher emotion recognition accuracy. In addition, its average recognition accuracy on each data set reaches 90%, indicating that the model is very suitable for the sentiment recognition of cross-border e-commerce reviews. Therefore, the proposed model can accurately judge the emotions of user evaluations, and help merchants make corresponding sales strategy adjustments, which has certain feasibility.