Enhancing products performance evaluation through hybrid DistilRoBERTa and BiGRU models.

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning (ML)
• Artificial Intelligence (AI)
• Consumer Behavior Analysis

Background:

• Online reviews significantly influence purchasing decisions, yet current sentiment analysis methods often neglect product complexity and usability.
• Existing approaches primarily focus on sentiment polarity, failing to capture crucial aspects like user-friendliness and functional effectiveness.
• There is a need for advanced analytical tools to interpret the nuanced information within customer reviews for better product evaluation.

Purpose of the Study:

• To develop a comprehensive framework for analyzing product reviews by integrating appraisal theory with hybrid deep learning models.
• To explore and quantify product complexity and usability from customer feedback using advanced NLP techniques.
• To provide explainable insights into user perceptions, moving beyond traditional sentiment analysis.

Main Methods:

• A hybrid deep learning model combining DistilRoBERTa (transformer) and bi-directional gated recurrent unit (RNN) was developed.
• Amazon product reviews were manually annotated for perceived complexity (appreciation, judgment) based on appraisal theory.
• The model utilized fine-tuned DistilRoBERTa embeddings and a bi-directional GRU layer, with 5-fold cross-validation and class weighting for imbalanced data.

Main Results:

• The proposed hybrid model achieved a mean accuracy of 96.13%, outperforming existing state-of-the-art methods like Random Forest, DistilBERT, XLNet, and GPT-based models.
• Shapley Additive Explanations (SHAP) were employed for model interpretability, revealing insights into emotional tendencies, functional effectiveness, and user perceptions.
• The model demonstrated superior performance in capturing contextual and sequential dependencies within product reviews.

Conclusions:

• The developed framework offers a scalable, automated solution for evaluating product performance and understanding user feedback on e-commerce platforms.
• Integrating appraisal theory with hybrid deep learning provides a more nuanced understanding of product reviews than traditional sentiment analysis.
• The explainable AI component enhances transparency, aiding in optimizing product development strategies and setting new standards for feedback analysis.