The mechanism that Airbnb always incorporates for 'smart pricing' setting a proper price

By Steel Wool

In Airbnb where you can share your own vacant room and get a usage fee, "host" which provides room can freely decide price. There is a mechanism "Smart pricing" for providing "appropriate price" according to various circumstances such as the time, the situation of the neighboring property, event information and the like, and the article " Customized Regression Model for Airbnb Dynamic Pricing (customized regression model for Airbnb's dynamic pricing) "has been announced.

KDD 2018 | Customized Regression Model for Airbnb Dynamic Pricing
http://www.kdd.org/kdd2018/accepted-papers/view/customized-regression-model-for-airbnb-dynamic-pricing

Customized regression model for Airbnb dynamic pricing | the morning paper
https://blog.acolyer.org/2018/10/03/customized-regression-model-for-airbnb-dynamic-pricing/

Airbnb is a service unique to the Internet society in which ordinary people offer prices to their vacant rooms with prices. However, "being able to decide the price freely" can be "double-edged sword", setting an appropriate price taking into consideration the situation of the property and competitiveness etc. for an amateur who is not an expert, maximizing profit It is not easy to do. If the profit is lowered too much or the price is raised too much to lower the occupancy rate, the total profit will be lowered and the quality of service will fall as Airbnb as a platform as well It is born unfavorable situation.

As a solution to solve such problems, Airbnb has a function called smart pricing that calculates a competitive overnight charge according to the demand in the area where the host provides the property and proposes the recommended fee It offers.

The recommended pricing for smart pricing is generated while comprehensively judging the host's desired conditions and a lot of data, and it is said that the factor determining factor actually reaches 70 or more. The following factors are reflected in price update.

Mechanism of 'Smart pricing' talked by people inside - The Airbnb Blog - Belong Anywhere
https://blog.atairbnb.com/smart-pricing-locale-en/

· Remaining time: As the check-in date approaches, the charge will change
· Popularity of the area: As the search area of ​​the whole area increases, the price changes
· Season: Charges change when entering busy season or quiet period
· Listing Popularity: As the number of views and reservations increases, the price will change
· Listing description information: If you increase the amenity and facilities such as Wi-Fi, the charge will change
· Reservation history: When the reservation is made, the charge at the contracting stage also affects the subsequent charge. For example, if you set a higher price manually than the recommended pricing for smart pricing, so when you make a reservation, the algorithm learns from it and reflects it on the recommended fee.
· Review history: Charges will change as more reviews are added

The price in smart pricing is set on the basis of the above factors, and further the price is adjusted according to the "Customized Regression Model" that reflects actual usage etc. It is said that it is. Although the mechanism is described in detail in the paper, understanding progresses if you look at the following movie made based on the paper before actually reading.

Customized Regression Model for Airbnb Dynamic Pricing - YouTube


Airbnb is a service that allows you to share your home's room with users and provide a variety of local experiences (experiences).

It is Airbnb's service to connect a host who wants to provide a landlord and charming property that you do not use, and a user who is looking for a room other than the hotel at your destination.

Even in such Airbnb, there is a mechanism for price determination "demand and supply". However, it is not easy for hosts who are not professional traders with many know-how and systems to maximize profits by setting appropriate and competitive prices. And for Airbnb it is important to improve the quality of the whole service including users and hosts by setting more attractive prices.

There, Airbnb offers several pricing tools. Airbnb provides information such as the optimal unit price of the property, weekend price, long-term discount price, etc. to the host, and by offering "recommended price" and "smart pricing" functions, it is "too low, too high It is aiming at realizing the best pricing of "without" and increasing the occupancy rate of the property and service.

Smart pricing takes into account parameters such as "number of people staying", "scheduled date of use", "reservation status of other properties", "highest quality of the listing that the host is issuing", and the algorithm automatically selects the optimum price I will present it.

The price of smart pricing is based on the binary evaluation model "Booking Probability Model" that evaluates the possibility of being booked on each day and the "Strategy Model" which utilizes the regression model for determining the daily price (Strategic model) "and finally the logic" Personalization "phase that adjusts the output of the above process to determine the proposed price.

Reservability model is used to determine demand curve in Airbnb. There is a gap between the predicted demand curve (red) and the actual demand curve (green), setting the price too high to generate opportunity losses, or conversely setting the profit ratio too low It will become low. By bringing this prediction and the actual gap as closely as possible, the benefit of the host is maximized.

However, in practice it is very difficult to determine the price based on the correct demand curve. When considering "optimized price", it is important to think about what "bad price setting is"?

When there was a day when the reservation did not enter (green), it means that the price was set too high than the demand. The background may include factors such as "There were no attractive events in the area on that day" or "Weather was bad", but when we considered all of them as "demand" , It is judged that the price set in the listing was too high for the optimum price derived from that demand.

Conversely, if the offer price is cheaper than the optimal price for the reservation (green) day, this means that Smart pricing proposed a price lower than the optimal price .

In order to fill this gap, Airbnb is trying to improve the accuracy of smart pricing in the future by using the loss function.

In the thesis, since the mechanism of this area is described in detail using complicated mathematical expressions, it should be very helpful especially for those who are interested in data science. There are also blogs that touched the contents in Japanese, so that is likely to be helpful as well.

【KDD2018】 I read the article "Customized Regression Model for Airbnb Dynamic Pricing" and summarized - Notepad of marketing scientist suffering in Minato-ku
https://honawork.hatenablog.com/entry/2018/08/24/181947

【Paper note: Airbnb Pricing Model】 Customized Regression Model for Airbnb Dynamic Pricing - Re: ML life starting from zero
https://tsunotsuno.hatenablog.com/entry/2018/09/24/085427