Estimation of the Investability of Real Estate Properties through Text Analysis

The Multiple Listing Service, commonly known as the MLS, is the singularly most important database where real estate agents and brokers list real estate properties for sale. It is common that agents include textual comments pertinent to the property. Although the information content of comments varies, it is usually expressed in good faith and in many cases is helpful in shedding light on the overall condition and the value of the property. Therefore, it seems reasonable that semantic text analysis would be useful to evaluate properties, or aspects thereof. As far as we're aware of, no methodology to effectively extract insight from the MLS textual portion exists. In this paper we demonstrate how textual descriptions may be exploited for property ranking. The proposed methodology, which combines supervised and unsupervised methods, identifies domain-specific concepts and combines their contributions to assign a score to a listing. We evaluate the proposed methods using both human evaluators and data-driven evaluation metrics on real datasets (complied from actual listings), and compare them to baseline approaches.