Technical Talks

View All

Real Estate Search Engine Enhancement Using AI/ML: 5 Examples

Volodymyr Kubytskyi Volodymyr Kubytskyi | Chief Innovation Officer | Flatfy

As a Real Estate aggregator, we deal with huge amounts of data: our crawler processes 6M ads/mo from 10k sites. In addition, the information on the source site is usually poor: many ads have no address, price, or description; but we can enrich this thanks to technology.

Taking noisy data, we make it great with machine learning.

In this talk, we will discuss 5 ways we use AI/ML to improve user search experience:

- Offering new ways of filtering. Our cutting-edge Convolutional Neural Networks make it possible to recognize images, detect watermarks and even evaluate apartment renovation quality.

- Boosting deduplication to 99% accuracy. We have developed our own technology for duplicates detection, which is now being patented, and we will explain how we have created it from scratch.

- Predicting price for poorly filled ads. Algorithms based on gradient boosting (GBM) over decision trees predict property price with 92% accuracy, and spam prevention systems work with 96% F1 score.

- Enriching ads with accurate address and parameters. RNNs help us determine crucial information that is often missing from the source ad, such as the apartment's exact address, area, room count, floor and other features.

- Enhancing images. We've constructed GANs that can remove almost any watermark or logo from images so users can properly view them.

Volodymyr Kubytskyi
Volodymyr Kubytskyi
Chief Innovation Officer | Flatfy

Volodymyr Kubytskyi has been working on creating ML-based algorithms in PropTech for the last 5 years. He is currently Chief Innovation Officer at Flatfy & LUN – a leading Ukrainian PropTech that develops online platforms for real estate search, now active in 30 countries.

Volodymyr got his Master's degree in Applied Mathematics. In his master thesis, he developed an image deduplication system based on CNNs, which is now integrated into the Flatfy workflow.His research interests include computer vision, deep learning, and gradient boosting algorithms with final application in products.