Under the hood

From time to time we are delighted to welcome guest authors to the Expedia Viewfinder blog. In this post, Jan Krasnodebski, Expedia’s director of lodging revenue optimization, gives us an inside-baseball look at the technology behind Expedia’s travel deals search engine.

At Expedia, one of our primary goals is to present travelers with the best offers and deals. It may seem like a simple thing to do, but there actually are a lot of factors to consider before we show a customer top search results.

Let’s look at a couple of different scenarios: business travelers versus a family, traveling with children. These two types of customers typically look for specific–and very different–qualities in a hotel. A business traveler may want to be close to a financial district, or he may want to use a hotel that will offer rewards, and has an upscale restaurant. This traveler also might be willing to pay for these extras, On the other side of the equation, a traveling family may be on a tighter budget and therefore may be searching for amenities such as a shuttle to and from the airport, access to an indoor pool, and proximity to local tourist attractions. These characteristics can be identified by analyzing hundreds of thousands of purchases.

Historically, when determining search results, we used classical statistics. Essentially, this approach makes an assumption about the search someone entered, then we build an algorithm to address that assumption, and test it and adjust it based on the outcome (whether or not the search result led to a hotel booking). In recent years, newer technological advancements have led to an alternative process called machine learning. This format lets the data do the talking: algorithms used to present search results adapt automatically to make a better prediction about what the customer is looking for. Essentially, the process learns from the data it is provided. The more data, the better the result, giving a leading OTA such as Expedia an advantage in understanding customer preferences.

These changes toward machine learning may not be evident to customer immediately; in reality, our hope is that our customers don’t notice a change at all, and simply are pleased by the hotel search criteria we present. On a practical level, one can see the differences by looking at what hotels customers buy. In general, customers buy the hotel we place in the first position more than 30 percent of the time, and line up with the top 5 more than 50 percent of the time.

By implementing machine learning, we’re able to better address our customers’ needs and offer them the best deals.

Analyzing this type of big data is called data science. Expedia.com recently participated in the International Conference on Data Mining (ICDM), where we were able to discuss some of these practices in great detail. Every year, ICDM hosts a competition on data mining, and this year Expedia was chosen to be the sponsor. We received more than 300 submissions for how to address hotel search queries, and the results were quite impressive. We’re pleased to have made some strong connections at the conference and are working with some of the winners on implementing some of the ideas discussed. We look forward to presenting more on our work in this space over the coming year. 

In the meantime, our goal is to deliver the best deals for our customer. We take a lot of information into account to present these deals, and we welcome feedback–let us know how we’re doing by responding in the comments of this post, or by connecting with us on Facebook or Twitter. Thanks in advance!