Unboxing Deep Learning Model of Food Delivery Service Reviews Using Explainable Artificial Intelligence (XAI)...

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Customer Service Analytics

Background:

• The surge in food delivery services (FDSs) during COVID-19 generated vast amounts of customer feedback.
• Analyzing this feedback manually is resource-intensive, hindering FDSs' ability to address customer concerns effectively.
• Deep learning (DL) offers potential for automated analysis, but its 'black box' nature requires explainability.

Purpose of the Study:

• To compare the performance of various DL models for sentiment analysis in the FDS domain.
• To utilize explainable AI (XAI) techniques to interpret DL model predictions and identify key sentiment drivers.
• To select an optimal DL model for FDSs prioritizing the accurate identification of customer complaints.

Main Methods:

• Sentiment analysis was performed using Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and a hybrid Bi-GRU-LSTM-CNN model.
• Models were trained and tested on customer reviews from the ProductReview website.
• Explainable AI (XAI) methods, specifically SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME), were employed for model interpretability.

Main Results:

• The DL models achieved high accuracy: LSTM (96.07%), Bi-LSTM (95.85%), and Bi-GRU-LSTM-CNN (96.33%).
• The LSTM model demonstrated the lowest rate of false negatives, crucial for FDSs aiming to address all customer complaints.
• XAI techniques successfully identified word-level contributions to sentiment, validating the models' predictions.

Conclusions:

• The LSTM model is a suitable choice for FDS sentiment analysis due to its high accuracy and low false negative rate.
• Integrating XAI with DL models enhances transparency and provides actionable insights for improving FDS customer satisfaction.
• This approach enables FDS organizations to efficiently process customer feedback and drive service enhancements.