A Deep Learning-Based Sentiment Classification Model for Real Online Consumption.

Area of Science:

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

Background:

• Online consumer reviews are crucial for purchasing decisions but often contain irrelevant or malicious content.
• Accurate sentiment analysis of online reviews is vital for businesses and consumers.
• Existing methods struggle to effectively discern genuine sentiment from noisy review data.

Purpose of the Study:

• To develop a deep learning model for accurate real online consumer sentiment classification.
• To enhance the reliability of sentiment analysis in e-commerce environments.
• To improve the timely identification of genuine consumer sentiment tendencies.

Main Methods:

• Established a mapping from product reviews to sentiment features using expert knowledge and fuzzy mathematics.
• Utilized convolutional operations for local contextual features and bidirectional long- and short-term memory networks for long-term dependencies.
• Incorporated an attention mechanism to weigh word contributions and a regular term constraint in the objective function.

Main Results:

• The proposed convolutional attention-long and short-term memory network (CA-LSTM) model achieved a test accuracy of 83.3%.
• The CA-LSTM model demonstrated superior classification performance compared to other benchmark models.
• The model effectively maps high-dimensional text data into a low-dimensional space for feature extraction.

Conclusions:

• The CA-LSTM model offers a robust solution for sentiment classification of online consumer reviews.
• Deep learning approaches, particularly with attention mechanisms, can significantly improve sentiment analysis accuracy.
• Accurate sentiment analysis aids in understanding consumer experiences and combating review manipulation in e-commerce.