A study of text classification algorithms for live-streaming e-commerce comments based on improved BERT model.

Area of Science:

• Natural Language Processing
• Machine Learning
• E-commerce Analytics

Background:

• E-commerce live streaming generates vast amounts of brief, diverse 'bullet comments'.
• Analyzing these comments is crucial for understanding customer engagement and marketing effectiveness.
• Existing methods struggle with the volume and complexity of bullet comment data.

Purpose of the Study:

• To develop an improved model for classifying e-commerce bullet comments.
• To enhance the accuracy and efficiency of bullet comment analysis.
• To facilitate valuable information extraction for marketing purposes.

Main Methods:

• Proposed an improved BERT model utilizing a hierarchical classification structure.
• Trained a parent class BERT model for broad categorization.
• Developed subclass BERT models for fine-grained classification within categories.

Main Results:

• The hierarchical BERT model significantly improved classification accuracy.
• The model demonstrated enhanced efficiency in processing large comment volumes.
• Empirical evidence confirmed the model's effectiveness.

Conclusions:

• The hierarchical BERT approach offers a robust solution for e-commerce bullet comment analysis.
• This method aids in extracting actionable insights from live stream interactions.
• Improved analysis supports more effective marketing strategies in e-commerce.