Abstract

The Australian housing market is one of the most dynamic and competitive in the world, consistently driven by the pursuit of the best possible selling outcomes. This study unpacks the context-specific sentiment embedded in property listing descriptions, introducing a novel approach to contextualised feature engineering and sentiment analysis that improves prediction of property selling outcomes. Drawing on domain knowledge, we construct three context-specific sentiments uniquely relevant to property sales: scarcity pressure, trust reassurance, and affordability concerns. Using advanced natural language processing techniques, we extract and quantify these features from listing texts. We demonstrate their effect by integrating them into machine learning models predicting property selling outcomes, where they significantly improve performance compared with general sentiment features. Beyond prediction, our findings show how contextualised sentiment provides a deeper understanding of the persuasive and affective signals within property listing descriptions, and offers practical insights into how descriptions shape market outcomes.

Download

Share

COinS