Using BigDL on Apache Spark to Improve the MLS Real Estate Search Experience at Scale

Download Slides

BigDL-enabled Deep Learning analysis of photos attached to property listings in Multiple Listings Services database allowed us to extract image features and identify similar-looking properties. We leveraged this information to in real-time property search application to improve the relevancy of user search results. Imagine identifying a property listing photo you like and having the system suggest other listings you should also review. Traditional real-estate MLS (multiple-listings services) search methods rely on SQL-type queries to search and serve real-estate listings results.

However, using BigDL in conjunction with MLSLinstings standard APIs allows users to include photos as search parameters in real-time, based both on image similarities and semantic feature search. The information extracted from listing’s images is used to improve the relevancy of the search results. To enable this use-case, we implemented several CNNs using BigDL framework on Microsoft’s Azure hosted Apache Spark: – Image feature extraction and tagging. Extracts features from real estate images and classifies them according to Real Estates Standards Organization rules, such as overall house style, interior and exterior attributes, etc. – Image similarity network which allows for comparing images that belong to different properties based on their extracted features and create a similarity score to be used in search results.

We’ll discuss the above networks in details as well as run a live demo of real-estate search results. Key takeaways: a) Why invest into Spark BigDL from the start. b) Why choose cloud-based solution from the start. c) Choice of Scala vs Python.

Session hashtag: #ExpSAIS16



« back
About Sergey Ermolin

Sergey is a technical program manager and Solutions Architect for deep learning, Spark analytics, and big data technologies at Intel. A Silicon Valley veteran with a passion for machine learning and artificial intelligence, Sergey's interest in neural networks goes back to 1996, when he used them to predict aging behavior of quartz crystals and cesium atomic clocks built by Hewlett-Packard. Prior to Intel, he held technical leadership positions at Apple and several Valley startups. Sergey holds an MSEE and a MSCS Mining Massive Datasets certificate from Stanford as well as BS degrees in Physics and Mechanical Engineering from CalState, Sacramento

About Dave Wetzel

As MLSListings’ Chief Operations Officer, David Wetzel is the day-to-day partner to MLSListings’ visionary CEO. David is responsible for overseeing the functional, service, and technology groups to ensure they are working cohesively to execute the company’s aggressive strategic plan. David is a veteran Silicon Valley software and technology executive. Prior to joining the MLSListings leadership team, David made his mark with companies such as IBM, Siemens, and Cisco Systems, in positions spanning Product Management, Marketing, Software Architecture, and Software Development. David holds a Bachelor of Science degree in Computer Science from California Polytechnic State University-San Luis Obispo, and a Masters in Engineering Management from Santa Clara University. Most recently, David added the prestigious CMLX3 (Certified Multiple Listings Expert) designation, which is the most comprehensive MLS industry leadership program ever offered and is reserved for those ready to make a lasting contribution to the marketplace.