At Airbnb, we endeavor to construct a world where anyone can belong anywhere. We try to know what our guests care about and match them with Hosts who can provide what they’re in search of. What higher source for guest preferences than the guests themselves?

We built a system called the (APS) to take heed to our guests’ needs in a house: What are they requesting in messages to Hosts? What are they commenting on in reviews? What are common requests when calling customer support? And the way does it differ by the house’s location, property type, price, in addition to guests’ travel needs?

With this personalized understanding of what home amenities, facilities, and site features (i.e. “home attributes”) matter most to our guests, we advise Hosts on which home attributes to accumulate, merchandize, and confirm. We may also display to guests the house attributes which can be most relevant to their destination and wishes.

We do that through a scalable, platformized, and data-driven engineering system. This blog post describes the science and engineering behind the system.

First, to find out what matters most to our guests in a house, we take a look at what guests request, comment on, and call customer support about probably the most. Are they asking a Host whether or not they have wifi, free parking, a non-public hot tub, or access to the beach?

To parse this unstructured data at scale, Airbnb built (isting tribute traction), a machine learning system that may extract home attributes from unstructured text data like guest messages and reviews, customer support tickets, and listing descriptions. LATEX accomplishes this in two steps:

1. A extracts key phrases from unstructured text data
2. An then maps these key phrases to home attributes

The named entity recognition (NER) module uses textCNN (convolutional neural network for text) and is trained and advantageous tuned on human labeled text data from various data sources inside Airbnb. Within the training dataset, we label each phrase that falls into the next five categories: Amenity, Activity, Event, Specific POI (i.e. “Lake Tahoe”), or generic POI (i.e. “post office”).

The entity mapping module uses an unsupervised learning approach to map these phrases to home attributes. To realize this, we compute the cosine distance between the candidate phrase and the attribute label within the fine-tuned word embedding space. We consider the closest mapping to be the referenced attribute, and may calculate a confidence rating for the mapping.

We then calculate how continuously an entity is referenced in each text source (i.e. messages, reviews, customer support tickets), and aggregate the normalized frequency across text sources. Home attributes with many mentions are considered more necessary.

With this technique, we’re in a position to gain insight into what guests are eager about, even highlighting recent entities that we may not yet support. The scalable engineering system also allows us to enhance the model by onboarding additional data sources and languages.

What guests search for in a mountain cabin is different from an urban apartment. Gaining a more complete understanding of guests’ needs in an Airbnb home enables us to supply more personalized guidance to Hosts.

To realize this, we calculate a singular rating of attributes for every home. Based on the characteristics of a house–location, property type, capability, luxury level, etc–we predict how continuously each attribute shall be mentioned in messages, reviews, and customer support tickets. We then use these predicted frequencies to calculate a customized importance rating that’s used to rank all possible attributes of a house.

For instance, allow us to consider a mountain cabin that may host six individuals with a mean day by day price of $50. In determining what’s most vital for potential guests, we learn from what’s most talked about for other homes that share these same characteristics. The result: hot tub, fire pit, lake view, mountain view, grill, and kayak. In contrast, what’s necessary for an urban apartment are: parking, restaurants, grocery stores, and subway stations.

We could directly aggregate the frequency of keyword usage amongst similar homes. But this approach would run into issues at scale; the cardinality of our home segments could grow exponentially large, with sparse data in very unique segments. As a substitute, we built an inference model that uses the raw keyword frequency data to infer the expected frequency for a segment. This inference approach is scalable as we use finer and more dimensions to characterize our homes. This permits us to support our Hosts to best highlight their unique and diverse collection of homes.

Now that we’ve got a granular understanding of what guests want, we may help Hosts showcase what guests are in search of by:

• Recommending that Hosts acquire an amenity guests often request (i.e. coffee maker)
• Merchandizing an existing home attribute that guests are likely to comment favorably on in reviews (i.e. patio)
• Clarifying popular facilities which will find yourself in requests to customer support (i.e. the privacy and talent to access a pool)

But to make these recommendations relevant, it’s not enough to know what guests want. We also must ensure about what’s already in the house. This seems to be trickier than asking the Host resulting from the 800+ home attributes we collect. Most Hosts aren’t able to right away and accurately add all the attributes their home has, especially since amenities like a crib mean various things to different people. To fill in a few of the gaps, we leverage guests feedback for amenities and facilities they’ve seen or used. As well as, some home attributes can be found from trustworthy third parties, corresponding to real estate or geolocation databases that may provide square footage, bedroom count, or if the house is overlooking a lake or beach. We’re in a position to construct a really complete picture of a house by leveraging data from our Hosts, guests, and trustworthy third parties.

We utilize several different models, including a Bayesian inference model that increases in confidence as more guests confirm that the house has an attribute. We also leverage a supervised neural network WiDeText machine learning model that uses features in regards to the home to predict the likelihood that the subsequent guest will confirm the attribute’s existence.

Along with our estimate of how necessary certain home attributes are for a house, and the likelihood that the house attribute already exists or needs clarification, we’re able to provide personalized and relevant recommendations to Hosts on what to accumulate, merchandize, and make clear when promoting their home on Airbnb.

That is the primary time we’ve known what attributes our guests want all the way down to the house level. What’s necessary varies greatly based on home location and trip type.

This full-stack prioritization system has allowed us to provide more relevant and personalized advice to Hosts, to merchandize what guests are in search of, and to accurately represent popular and contentious attributes. When Hosts accurately describe their homes and highlight what guests care about, guests can find their perfect vacation home more easily.

We’re currently experimenting with highlighting amenities which can be most vital for every form of home (i.e. kayak for mountain cabin, parking for urban apartment) on the house’s product description page. We consider we will leverage the knowledge gained to enhance search and to find out which home attributes are most vital for various categories of homes.

On the Host side, we’re expanding this prioritization methodology to encompass additional suggestions and insights into how Hosts could make their listings much more desirable. This includes actions like freeing up popular nights, offering discounts, and adjusting settings. By leveraging unstructured text data to assist guests connect with their perfect Host and residential, we hope to foster a world where anyone can belong anywhere.

If such a work interests you, take a look at a few of our related positions at Careers at Airbnb!

It takes a village to construct such a strong full-stack platform. Special because of (alphabetical by last name) Usman Abbasi, Dean Chen, Guillaume Guy, Noah Hendrix, Hongwei Li, Xiao Li, Sara Liu, Qianru Ma, Dan Nguyen, Martin Nguyen, Brennan Polley, Federico Ponte, Jose Rodriguez, Peng Wang, Rongru Yan, Meng Yu, Lu Zhang for his or her contributions, dedication, expertise, and thoughtfulness!