Emotion analysis of user comments on e-commerce platforms based on deep learning

Sentiment analysis on millions of reviews is now a significant problem with an exponential growth in the number of user-generated content on e-commerce sites. The paper introduces a unique emotion recognition methodology that aims at addressing these issues through a hybrid deep learning methodology. The model involves the use of a transformer-based architecture that uses contrastive learning and Battlefield Optimization Algorithm (BFOA) to optimize the model. The approach is meant to solve certain issues, including sarcasm, domainspecific language, and imbalance in the number of classes, which are common in sentiment analysis with e-commerce. A large set of ecommerce reviews was tested using the model. It reached the precision of 98.65, 98.67, the recall of 98.65 and the F1-score of 98.64. During the testing, the model has demonstrated low false positive and false negative rates, hence its strength. The proposed model was found to be much better in the accuracy and performance of overall classification as compared to the existing methods of sentiment analysis which were used to carry out comparative experimental works. Such findings can guarantee the potential of the proposed model to enhance the sentiment analysis systems on the e-commerce site towards oriented insights into business decisions.