The popular online housing platform, Airbnb, has expanded into services such as touring, hiking, and all sorts of traveling events. While these offerings are fantastic for customers, the company is looking to improve on them even more. To do so, Airbnb wants to use artificial intelligence.
As reported by VentureBeat, a paper was published to Arxiv.org, entitled “Applying Deep Learning To Airbnb Search.” In it, the group reveals their plans over the past two years regarding the emerging technology. Now, they’re working on a neural network to break into a new model of ranking travel results. The research will apply to both the web and mobile applications to enhance the user experience.
The document reveals their reasoning:
“The application to search ranking is one of the biggest machine learning success stories at Airbnb. Much of the initial gains were driven by a gradient boosted decision tree model. The gains, how- ever, plateaued over time. This paper discusses the work done in applying neural networks in an attempt to break out of that plateau.”
As with most similar apps, users type their desired result in a search bar. However, the listing would bring in relevant outcomes that had taken from millions within Airbnb’s database. Houses and areas within the specified location would show, but the results were not the best they could be.
Traditionally, listings would rank based on a “manually crafted” method of scoring. Then, a “gradient boosted decision tree” (GBDT) would check and predict important information on top of this scoring section. At the time of this invention, the team said it is “one of the largest step improvements in home bookings in Airbnb’s history.”
But, now that these search results have lost their lust, the group is looking into A.I. machine learning.
Improving On The Foundation
However, the network doesn’t utilize just one A.I. system. Instead, it uses a myriad of different algorithms that list results based on the potential connection between the guest and the host. This system looks into how guests tip, their ratings of past experiences, and what the host is looking for. As users interact more with the network, it learns what people like, and results are more likely to become a match.
While the first system was the foundation, the second utilized a method called LambdaRank. This process watched over the machine learning to solve any problems, while the third was the deep neural network (DNN) detailed in the paper.
The DNN utilized 195 different attributes such as price, offered services, ratings, ones that use Airbnb’s “Smart Pricing” tool, and recommended spaces based on previous rentals.
Of course, the artificial intelligence platform wasn’t perfect at first. There were different methods before the current one. One provided each listing with their own ID. Then, the database uses that ID to organize each unique aspect of an offering. However, this proved to be too much information for any efficiency.
A closer model resulted in basing search results on the length of viewing a listing. However, researchers found that a long look at a listing doesn’t always translate over to a booking. Sometimes users would stay on an offering for a while due to an odd description or outstanding pricing methods.
It was a tough system to crack, but the team believes the effort has lead to a fantastic platform:
“Feeding on the ubiquitous deep learning success stories, we started at the peak of optimism, thinking deep learning would be a drop in replacement for the GBDT model and give us stupendous gains out of the box,” the researchers wrote. “A lot of initial discussions centered around keeping everything else invariant and replacing the current model with a neural network to see what gains we could get … Over time we realized that moving to deep learning is not a drop-in model replacement at all; rather it’s about scaling the system. As a result, it required rethinking the entire system surrounding the model.”