Authors - Santhi Bharath Punati, Venkata Akhil Kumar Gummadi, Sandeep Kanta, Praveen Damacharla Abstract - The rise of global e-commerce demands accurate sentiment analysis across multiple languages to enhance customer experience and decision-making. However, existing sentiment analysis models struggle with multilingual and code-mixed data, leading to inconsistencies in customer sentiment interpretation. This research presents an advanced deep learning framework that integrates Multilingual BERT (mBERT) embeddings with an Attention-Augmented Gated Recurrent Unit (GRU) network to improve sentiment classification across diverse linguistic contexts. A dataset of 13,000 customer reviews spanning English, Hindi, Hinglish, German, and Spanish was processed using mBERT for contextual embedding, addressing tokenization and syntactic variability challenges. The proposed hybrid model leverages transformer-based contextual understanding with the sequence modeling capabilities of GRU, while the attention mechanism enhances key sentiment features. Experimental evaluations demonstrate the superiority of our model, achieving 93.45% test accuracy and a test loss of 0.0974, outperforming conventional architectures such as LSTM, BiLSTM, and BiLSTM-GRU. The results confirm the model’s effectiveness in maintaining contextual integrity and sentiment accuracy across multilingual datasets. This framework offers a scalable and adaptable solution for e-commerce platforms, enabling businesses to derive precise sentiment insights from global customer reviews. By addressing challenges in multilingual sentiment analysis, our approach facilitates personalized customer engagement, improved product recommendations, and strategic business decisions. Future research may explore expanding sentiment analysis to low-resource languages and real-time feedback systems, further strengthening the inclusivity and intelligence of e-commerce analytics.
Open Zoom
