Sentiment analysis for e-commerce product reviews by deep learning model of Bert-BiGRU-Softmax

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Machine Learning

Background:

• Sentiment analysis of e-commerce reviews is crucial for product quality management.
• Manufacturers and customers benefit from understanding public sentiment towards products.

Purpose of the Study:

• To propose a novel deep learning model for accurate e-commerce review sentiment analysis.
• To improve the extraction of multi-dimensional product features and sentiment nuances from reviews.

Main Methods:

• Developed a Bert-BiGRU-Softmax model with hybrid masking, review extraction, and attention mechanisms.
• Utilized the BERT model for feature extraction, Bidirectional GRU for semantic coding and sentiment weighting, and Softmax with attention for classification.
• Conducted experiments on a large-scale dataset of over 500,000 e-commerce reviews.

Main Results:

• The proposed Bert-BiGRU-Softmax model achieved over 95.5% accuracy.
• The model demonstrated superior performance compared to other deep learning models like RNN, BiGRU, and Bert-BiLSTM.
• The model effectively retained a lower loss in sentiment classification tasks.

Conclusions:

• The novel deep learning model significantly enhances sentiment analysis accuracy for e-commerce reviews.
• This approach offers a valuable tool for manufacturers to gauge public sentiment and improve product quality.
• The model provides a more nuanced understanding of customer attitudes towards products.