INTEGRATING EXPLAINABLE AI (XAI) INTO CHATBOTS FOR AUTOMATED DECISION-MAKING IN HOSPITALITY

Abstract

This paper presents the design and implementation of a hybrid generative AI chatbot for the hospitality domain, integrating Explainable Artificial Intelligence (XAI) techniques to enhance transparency, build trust, and support decisionmaking. The system processes guest data to generate responses with suggestions and determine appropriate actions, including the necessity in offering upgrades, providing additional details or delivering appreciation messages to improve guest experience. The system combines multiple technologies starting with Pandas for structured data handling, model RandomForestClassifier for user classification, SHapley Additive exPlanation (SHAP) as an interpreter of forecasting model, NumPy for numerical operations and OpenAI GPT-3.5 API for natural language generation based on the request. To identify most influential features in the decision-making process was used SHAP. Prompt engineering was used to ensure accuracy and professionalism in responses. The evaluation of the model includes classification reports with accuracy, recall, and F1-score metrics. The system demonstrates the alignment between AI-driven predictions with real world operations and personalized responses based on contextual and probabilistic factors. By integrating XAI with generative models the aim is to improve both operational efficiency and customer feedback. This integration results in a combination of automated context-aware responses and transparent decision explanations. Results were interesting after evaluating each class of the dataset. Finally, this study serves as an aid to increase trust and confidence in guests, mainly in explaining the answer provided by the chatbot as well as future research directions involving alternative algorithms, larger datasets and optimization strategies.