Enhancing Customer Experience in E-commerce through Multilingual Sentiment Analysis

The study presents the novel framework of examining sentiments in more than one language that intends to increase the rate of customer satisfaction on the international ecommerce portals that are Amazon, Walmart, Apple, eBay, and Etsy. The objective of multilingual BERT (mBERT) was to work on a dataset containing 13,000 customer reviews in English and Hindi/English and English/Hindi code-mixing and German/Spanish content. It then encoded and coded the texts that would be used later on. We designed a unified architecture of mBERT embeddings and an Attention-Augmented BiLSTM-GRU hybrid layer that enables it to address linguistic peculiarities and situational ambiguities. The suggested approach relates to understanding the context of the whole world by using transformers, and constructs GRUs that are trained on association modeling with some attention mechanics to be used in identifying important aspects of texts. The given model attained 93.45 percent test accuracy and 0.0974 as test loss that established better performance over regular architectures including LSTM, BiLSTM and BiLSTM-GRU. The framework provides a flexible road map on how to optimize sentiment analysis in broad linguistic contexts by eliminating tokenizing bugs in the existing systems as it incorporates sophisticated deep learning algorithms that further maximize customer satisfaction rate.