The language of neighborhoods: A predictive-analytical framework based on property advertisement text and mortgage lending data

Abstract

Real estate property listings use specific language to market properties to a target buyer – typically one that will garner the largest profit. As home-seekers have different preferences for house characteristics and neighborhood amenities, the words used to advertise homes are expected to vary according to the type of neighborhood and expected homebuyer. In this article, we develop a framework for extracting the key characteristics used to advertise properties according to the racial and income profile of home mortgage applicants in different types of neighborhoods. We perform an exploratory text analysis on words according to neighborhood types and use a binomial logistic regression model to determine the most discriminatory words for each type of neighborhood. Finally, we assess the ability of the property listing text to predict the type of neighborhood the property belongs to. Using a small, illustrative case study of listings from Charlotte, North Carolina, we find that the presence of specific neighborhood names holds more importance in neighborhoods with primarily White homebuyers. In gentrifying neighborhoods, unique property characteristics such as parquet flooring, and words associated with revitalization near the city center are common. Listings in neighborhoods with minority homebuyers are less likely to mention schools and feature traditionally suburban descriptors such as cars, garage, and roadways. We envision that this framework, using near real-time data sources, holds the potential to advance neighborhood prediction efforts, our understanding of amenity preferences and sorting patterns, and to illuminate less visible processes of change such as discrimination in the housing market.