Application of an emotional classification model in e-commerce text based on an improved transformer model

Area of Science:

• Natural Language Processing (NLP)
• Deep Learning
• Machine Learning

Background:

• Increasing reliance on online reviews for product and store feedback.
• Text sentiment classification is a key research area for analyzing user opinions.
• Existing deep learning methods face challenges like information loss and weak context.

Purpose of the Study:

• To improve text sentiment classification accuracy and recall for e-commerce reviews.
• To address limitations of traditional deep learning models like CNN and RNN.
• To reduce model training difficulty and time costs.

Main Methods:

• An improved transformer model, replacing traditional CNN and RNN architectures.
• Application of IN standardization and GELU activation functions for enhanced analysis.
• Preprocessing steps including word vectorization for e-commerce review data.

Main Results:

• Significant improvements in recall (9.71%, 6.05%, 5.58%, 5.12%) compared to baseline models.
• Achieved peak F1 score levels in the test model.
• Demonstrated superior performance over BiLSTM, Naive Bayesian, BiLSTM_CNN, and attention-based BiLSTM models.

Conclusions:

• The proposed enhanced transformer model effectively improves text sentiment classification accuracy.
• The method offers a more efficient and accurate approach for analyzing user emotions in online reviews.
• Validated applicability of the method for general text classification tasks.