Beyond Sentiment: Context-Aware Emotion Detection in Guest Feedback for Intelligent Hospitality Systems

Abstract

The hospitality industry increasingly relies on artificial intelligence to enhance guest experiences through intelligent feedback analysis. Traditional sentiment analysis approaches fall short in capturing the nuanced emotional states expressed in guest reviews and feedback. This paper presents a novel context-aware emotion detection framework that transcends binary sentiment classification by identifying eight primary emotions—joy, trust, fear, surprise, sadness, disgust, anger, and anticipation—within hospitality guest feedback. Our methodology integrates transformer-based language models with contextual embeddings to achieve 87.3% accuracy in emotion classification. Through analysis of 50,000 hotel reviews, we demonstrate that context-aware emotion detection provides actionable insights for personalized service recovery and proactive guest experience management. The findings reveal that 34% of seemingly positive reviews contain underlying negative emotions requiring immediate attention, highlighting the limitations of conventional sentiment analysis in hospitality applications. Benchmarking against recent transformer-based emotion recognition models — RoBERTa, DeBERTa-v3, DistilBERT, EmoBERTa, SpanEmo and a zero-shot large language model baseline — confirms that domain-adapted, aspect-aware fine-tuning yields consistent gains in micro-F1 and macro-F1 over general-purpose architectures. We further report the preprocessing protocol in full and analyse latency, integration, monitoring and governance requirements for deploying the framework within operational hospitality systems.