"SeoulHouse2Vec": An Embedding-Based Collaborative Filtering Housing Recommender System for Analyzing Housing Preference

Abstract

Housing preference is the subjective and relative preference of users toward housing alternatives and studies in the field have been conducted to analyze the housing preferences of groups with sharing the same socio-demographic attributes. However, previous studies may not suggest the preference of individuals. In this regard, this study proposes "SeoulHouse2Vec," an embedding-based collaborative filtering housing recommendation system for analyzing atypical and nonlinear housing preference of individuals. The model maps users and items in each dense vector space which are called embedding layers. This model may reflect trade-offs between the alternatives and recommend unexpected housing items and thus improve rational housing decision-making. The model expanded the search scope of housing alternatives to the entire city of Seoul utilizing public big data and GIS data. The preferences derived from the results can be used by suppliers, individual investors, and policymakers. Especially for architects, the architectural planning and design process will reflect users' perspective and preferences, and provide quantitative data in the housing decision-making process for urban planning and administrative units.