Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Location Matters: Leveraging Multi-Resolution Geo-Embeddings for Housing Search

Ivo Silva, Pedro Nogueira, Guilherme Bonaldo

TL;DR

The study tackles location-driven housing recommendations by addressing item sparsity and broad location preferences with a geo-aware multi-resolution embedding framework. It introduces a two-tower architecture augmented with hierarchical $H3$ embeddings across resolutions $6$–$9$ and city context, trained via a contrastive InfoNCE objective. Intrinsic metrics (Information Abundance and spectral decay) and extrinsic production-log metrics (rent-flow uplift) show substantial improvements over Matrix Factorization and single-resolution baselines, with multi-resolution embeddings delivering the strongest gains. These findings imply meaningful enhancements to location-aware recommendations on large rental platforms and motivate online validation and richer geographic attributes in future work.

Abstract

QuintoAndar Group is Latin America's largest housing platform, revolutionizing property rentals and sales. Headquartered in Brazil, it simplifies the housing process by eliminating paperwork and enhancing accessibility for tenants, buyers, and landlords. With thousands of houses available for each city, users struggle to find the ideal home. In this context, location plays a pivotal role, as it significantly influences property value, access to amenities, and life quality. A great location can make even a modest home highly desirable. Therefore, incorporating location into recommendations is essential for their effectiveness. We propose a geo-aware embedding framework to address sparsity and spatial nuances in housing recommendations on digital rental platforms. Our approach integrates an hierarchical H3 grid at multiple levels into a two-tower neural architecture. We compare our method with a traditional matrix factorization baseline and a single-resolution variant using interaction data from our platform. Embedding specific evaluation reveals richer and more balanced embedding representations, while offline ranking simulations demonstrate a substantial uplift in recommendation quality.

Location Matters: Leveraging Multi-Resolution Geo-Embeddings for Housing Search

TL;DR

The study tackles location-driven housing recommendations by addressing item sparsity and broad location preferences with a geo-aware multi-resolution embedding framework. It introduces a two-tower architecture augmented with hierarchical embeddings across resolutions – and city context, trained via a contrastive InfoNCE objective. Intrinsic metrics (Information Abundance and spectral decay) and extrinsic production-log metrics (rent-flow uplift) show substantial improvements over Matrix Factorization and single-resolution baselines, with multi-resolution embeddings delivering the strongest gains. These findings imply meaningful enhancements to location-aware recommendations on large rental platforms and motivate online validation and richer geographic attributes in future work.

Abstract

QuintoAndar Group is Latin America's largest housing platform, revolutionizing property rentals and sales. Headquartered in Brazil, it simplifies the housing process by eliminating paperwork and enhancing accessibility for tenants, buyers, and landlords. With thousands of houses available for each city, users struggle to find the ideal home. In this context, location plays a pivotal role, as it significantly influences property value, access to amenities, and life quality. A great location can make even a modest home highly desirable. Therefore, incorporating location into recommendations is essential for their effectiveness. We propose a geo-aware embedding framework to address sparsity and spatial nuances in housing recommendations on digital rental platforms. Our approach integrates an hierarchical H3 grid at multiple levels into a two-tower neural architecture. We compare our method with a traditional matrix factorization baseline and a single-resolution variant using interaction data from our platform. Embedding specific evaluation reveals richer and more balanced embedding representations, while offline ranking simulations demonstrate a substantial uplift in recommendation quality.

Paper Structure

This paper contains 15 sections, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Data and Features
  4. Modeling Approaches
  5. Matrix Factorization
  6. Two-tower Model (H3 Resolution 9)
  7. Two-tower Multi-Resolution Model
  8. Training Objective and Loss Function
  9. Implementation Details
  10. Evaluation
  11. Intrinsic Evaluation
  12. Extrinsic Evaluation
  13. Results and Discussion
  14. Conclusion
  15. Authors

Figures (3)

  • Figure 1: Two-tower multi-resolution model architecture.
  • Figure 2: Logarithm of singular values ($\sigma_1$, sorted descending) for each model’s normalized embedding covariance matrix.
  • Figure 3: Simulated Rent-Flow Improvement. Percentage uplift in average rental-flow events (visits + offers) when re-ranking with each geo-aware model.