Beyond the Noise: Uncovering the Value of Inconsistent Signals in User-Generated Content for Hotel Performance Forecasting

Abstract

Contemporary hospitality forecasting exhibits a dual tendency: methodologies overemphasize predictive accuracy while feature selection overlooks non-consensus information. Grounded in signaling theory, this paper conceptualizes signal inconsistency in user-generated content (UGC) into three categories—consumer perceptual difference, signal content deviation, and environmental concept drift—and introduces a three-dimensional prediction framework to unlock their predictive power for hotel profit and quality performance. The framework evaluates model performance, feature contribution, and effect direction by integrating natural language processing, ensemble learning, and explainable artificial intelligence. Results demonstrate that inconsistent signals constitute valuable predictors that capture hotels’ performance dynamics more effectively than some traditional UGC-based features, with certain features exhibiting heterogeneous predictive effects across profit and quality performance. The three-dimensional prediction framework reconciles the competing objectives of predictive accuracy and interpretability, providing practitioners with actionable tools for dynamic market monitoring and risk assessment through systematic UGC analysis.