Presented By O'Reilly and Cloudera
Make Data Work
Feb 17–20, 2015 • San Jose, CA

Inspiring Travel With Location Personalization and Experiment Design

Lu Cheng (Airbnb), Lisa Qian (Airbnb)
2:20pm–2:40pm Thursday, 02/19/2015
Business & Industry
Location: LL20 BC
Average rating: ****.
(4.71, 7 ratings)

At Airbnb, our current search product easily allows users to search for a specific location and connect with a local host. We will outline how we built a new personalized discovery product, Discovery, for those travellers who don’t know where they want to go. By showing users relevant places to visit and surfacing reasons why, we are able to inspire even more trips.

We will start by discussing how we used our unique dataset to glean worldwide travel patterns and better understanding of what motivates people to travel to a destination. We’ve built a classifier to mine location tags from our descriptions and reviews to understand patterns like people go to Napa for its wineries, Europe has some of the most romantic cities, and the West Coast quantifiably has some of the best hiking spots in the world. We’ll also discuss how we analyzed our aggregate guest search behavior to systematically understand what granularity to recommend a location at, i.e. whether our guest understands the city Kings Beach or the larger entity Lake Tahoe. Bringing all of this together, we’ll demonstrate how our deeper understanding of travel destinations delivers a highly personalized experience.

Measuring the impact of Discovery within Airbnb is extremely complex for several reasons:

  • Visitors do not need to log in to search or to browse listings, so someone who browses on multiple devices could potentially see multiple treatment groups.
  • The process of coming to airbnb.com and booking a listing can take several days. Our goal is to make Airbnb top of mind when booking travel, not necessarily to compel users to return to the site every day.

The session will cover these issues and the multiple ways of evaluating the Discovery experience. We’ll discuss how to rethink engagement and click through rates, and how to better A/B test in the context of these challenges. We’ll also describe the extent of the mixed group problem, how it can introduce bias to experiments, and the innovative ways we correct for it.

Lu Cheng

Airbnb

Lu Cheng recently graduated from UC Berkeley with a B.S. in EECS and has been a software engineer on the Airbnb Search & Discover team since February 2014. Since starting at Airbnb, Lu has been focusing on user personalization and building models to understand unique characteristics of locations.
Photo of Lisa Qian

Lisa Qian

Airbnb

Lisa is a data scientist at Airbnb, where she focuses on search and discovery. Prior to joining Airbnb, Lisa completed a PhD in Applied Physics at Stanford University. Outside of data science, Lisa enjoys playing the violin and road biking.