"Everywhere at Home": What technologies does Airbnb use?
Airbnb was founded in 2008 in San Francisco. Brian Cesky and Joe Gebbia, the founders of the service, barely made ends meet, and in order to scrape money for rent, they allowed strangers to sleep on inflatable mattresses in their apartment.
One year later, Airbnb received $ 7 million and reported about 700 thousand housing booking operations. The service continued its intensive development and today offers more than 2,000,000 houses, apartments or rooms in 192 countries and 33,000 cities.
However, to manage such a huge system and to contact a large number of clients and homeowners, it is difficult to do without the use of smart algorithms and computerized systems . This topic seemed interesting to us, because we prepared a material about which algorithms and techniques make the life of the company and its customers easier.
')
/ photo Lukas CC
One of the questions that Airbnb wants to answer is, “How do you know what is where you have never been before?”. At the moment, Airbnb answer this question as follows. The company's engineers have implemented a so-called model of determining significant places, which informs customers about points of interest located near the specified point.
And in order to find out what is so interesting in these places, the company offers to use a guide on the districts, which allows you to find out some interesting facts about them and see professional photos. Moreover, tourists can get recommendations from homeowners themselves, using the guidebooks written by them.
These services, the creation of which takes into account the opinion of the company's customers, help build a route and visit all the places that a particular city is famous for. However, not all points are marked on maps. Often, the inhabitants “form” the sights themselves, denoting areas, for example, connected by a common history.
The search for such places is the team of Zack Walker (Zack Walker), exploring the historical and current data, discussions on forums and letters from home owners. For example, if you are looking for a “wine district” in Northern California, then the service will provide either a Napa county or Sonoma county with lists of differences and benefits.
To simplify the work of applying such "popular" areas to the map, the company uses the AT-AT system, named after the All-Terrain Armored Transport - the walking tank from the Star Wars universe. It allows you to manually set polygons of different shapes on the map, denoting in their properties parameters such as continent, country, region, city, etc.
However, it is not always convenient and efficient to manage the map manually, because Airbnb has created software for automatically marking maps by region using machine learning technologies. The advantage of such a system is that it works faster than the manual version, because it is able to quickly mark out areas for new markets.
To do this, use a specially written library aerosolve . It is well suited for working with discharged interpreted features that are often used in searches (for example, keywords, filters). Library sources, applications, as well as examples, you can find at the link to GitHub.
Airbnb uses dynamic pricing algorithms, when indicative prices are recalculated daily based on current market conditions. The algorithm is set up in such a way as to take into account the presence of the special properties of the proposed housing.
The proposed housing within the same city is “scattered” in very different areas, and other positions in the list may turn out to be cottages, huts, castles or yurts, therefore the company's algorithms take into account three basic types of data: similarity, novelty and location.
To determine the similarities, quantitative parameters are collected and analyzed: how many people can accommodate housing, whether the whole object is surrendered, the number of reviews, etc. After that, an assessment is made of how many guests are willing to pay for specific positions.
The novelty parameter is needed only because everything changes too quickly in the tourism business, and when assessing the location of housing, schematic maps of neighborhoods and districts in major cities around the world are prepared taking into account local conditions. For example, in London, the cost of apartments in the Greenwich region may be twice the cost of housing in the docks area, located on the other side of the river. This approach allows you to accurately determine the cost of positions in the database, taking into account geographical features and structures: rivers, lakes, railway tracks and others.
Moreover, the company switched to dynamic pricing by choosing a machine learning model called a classifier. It analyzes all the parameters of positions in the database based on current market requirements, and then predicts how successful this or that housing will be.
The system calculates the recommended cost based on hundreds of parameters like breakfast included in the price, availability of a bath and even local holidays. For example, the annual SXSW festival is held in the city of Austin; therefore, during its days the company expects an increase in the number of offers from homeowners in the area.
Since Airbnb operates in 190 countries, companies have to maintain a huge amount of currencies and processors. Most of the time, systems function without problems, but sometimes you have to deal with situations where, for example, a certain currency cannot be processed.
In order to catch these deviations as quickly as possible, the company's team has created a real-time anomaly detection system. This allows you to track errors right at the time of A / B testing of new payment methods or during the launch of new products.
The main task of the anomaly detection system is to search for outliers in a sample of time series. For this, a least squares regression is performed with the introduction of dummy variables.
This model is quite simple, and, as a rule, it defines the current trend quite accurately, but in this case it is assumed that the time series is set for weekly seasonality - the model will not work with products oriented to other seasonal patterns.
If you want to build a model of time series, taking into account both the trend and seasonality, then a different model is used. To show how it works, an example of computer mice and keyboards is given in the company's blog - the author conducts a simulation of seasonality and trends, tries to identify anomalies.
Such a large company as Airbnb simply cannot do without the constant improvement of technologies and algorithms for predicting user desires and resource allocation. Maths and machine learning come to the rescue in these issues - with their help, the service is able to do more, providing a high level of service and convenience to customers.
PS Additional reading:
One year later, Airbnb received $ 7 million and reported about 700 thousand housing booking operations. The service continued its intensive development and today offers more than 2,000,000 houses, apartments or rooms in 192 countries and 33,000 cities.
However, to manage such a huge system and to contact a large number of clients and homeowners, it is difficult to do without the use of smart algorithms and computerized systems . This topic seemed interesting to us, because we prepared a material about which algorithms and techniques make the life of the company and its customers easier.
')
/ photo Lukas CC
Geographic modeling
One of the questions that Airbnb wants to answer is, “How do you know what is where you have never been before?”. At the moment, Airbnb answer this question as follows. The company's engineers have implemented a so-called model of determining significant places, which informs customers about points of interest located near the specified point.
And in order to find out what is so interesting in these places, the company offers to use a guide on the districts, which allows you to find out some interesting facts about them and see professional photos. Moreover, tourists can get recommendations from homeowners themselves, using the guidebooks written by them.
These services, the creation of which takes into account the opinion of the company's customers, help build a route and visit all the places that a particular city is famous for. However, not all points are marked on maps. Often, the inhabitants “form” the sights themselves, denoting areas, for example, connected by a common history.
The search for such places is the team of Zack Walker (Zack Walker), exploring the historical and current data, discussions on forums and letters from home owners. For example, if you are looking for a “wine district” in Northern California, then the service will provide either a Napa county or Sonoma county with lists of differences and benefits.
To simplify the work of applying such "popular" areas to the map, the company uses the AT-AT system, named after the All-Terrain Armored Transport - the walking tank from the Star Wars universe. It allows you to manually set polygons of different shapes on the map, denoting in their properties parameters such as continent, country, region, city, etc.
However, it is not always convenient and efficient to manage the map manually, because Airbnb has created software for automatically marking maps by region using machine learning technologies. The advantage of such a system is that it works faster than the manual version, because it is able to quickly mark out areas for new markets.
To do this, use a specially written library aerosolve . It is well suited for working with discharged interpreted features that are often used in searches (for example, keywords, filters). Library sources, applications, as well as examples, you can find at the link to GitHub.
Prediction of demand and pricing
Airbnb uses dynamic pricing algorithms, when indicative prices are recalculated daily based on current market conditions. The algorithm is set up in such a way as to take into account the presence of the special properties of the proposed housing.
The proposed housing within the same city is “scattered” in very different areas, and other positions in the list may turn out to be cottages, huts, castles or yurts, therefore the company's algorithms take into account three basic types of data: similarity, novelty and location.
To determine the similarities, quantitative parameters are collected and analyzed: how many people can accommodate housing, whether the whole object is surrendered, the number of reviews, etc. After that, an assessment is made of how many guests are willing to pay for specific positions.
The novelty parameter is needed only because everything changes too quickly in the tourism business, and when assessing the location of housing, schematic maps of neighborhoods and districts in major cities around the world are prepared taking into account local conditions. For example, in London, the cost of apartments in the Greenwich region may be twice the cost of housing in the docks area, located on the other side of the river. This approach allows you to accurately determine the cost of positions in the database, taking into account geographical features and structures: rivers, lakes, railway tracks and others.
Moreover, the company switched to dynamic pricing by choosing a machine learning model called a classifier. It analyzes all the parameters of positions in the database based on current market requirements, and then predicts how successful this or that housing will be.
The system calculates the recommended cost based on hundreds of parameters like breakfast included in the price, availability of a bath and even local holidays. For example, the annual SXSW festival is held in the city of Austin; therefore, during its days the company expects an increase in the number of offers from homeowners in the area.
Detection of anomalies in the payment system
Since Airbnb operates in 190 countries, companies have to maintain a huge amount of currencies and processors. Most of the time, systems function without problems, but sometimes you have to deal with situations where, for example, a certain currency cannot be processed.
In order to catch these deviations as quickly as possible, the company's team has created a real-time anomaly detection system. This allows you to track errors right at the time of A / B testing of new payment methods or during the launch of new products.
The main task of the anomaly detection system is to search for outliers in a sample of time series. For this, a least squares regression is performed with the introduction of dummy variables.
This model is quite simple, and, as a rule, it defines the current trend quite accurately, but in this case it is assumed that the time series is set for weekly seasonality - the model will not work with products oriented to other seasonal patterns.
If you want to build a model of time series, taking into account both the trend and seasonality, then a different model is used. To show how it works, an example of computer mice and keyboards is given in the company's blog - the author conducts a simulation of seasonality and trends, tries to identify anomalies.
Such a large company as Airbnb simply cannot do without the constant improvement of technologies and algorithms for predicting user desires and resource allocation. Maths and machine learning come to the rescue in these issues - with their help, the service is able to do more, providing a high level of service and convenience to customers.
PS Additional reading:
Source: https://habr.com/ru/post/312132/
All Articles