In this assignment:
-Perform a Scenario Design analysis https://www.booking.com/.
-Consider whether it makes sense for this recommender system to perform scenario design twice, once for the organization (e.g. Amazon.com) and once for the organization’s customers.
-Attempt to reverse engineer what you can about the site, from the site interface and any available information that you can find on the Internet or elsewhere.
-Include specific recommendations about how to improve the site’s recommendation capabilities going forward.
1-Who are your target users?
The target users are property (apartments, hotels, vacation home, or bed & breakfasts) guests booking their stay through this website.
2-What are their key goals?
Some of the key goals of property guests are to book their stay according to their preferences for property type, dates, price, location, number of persons to be accommodated, and property features (ie., wifi, pets accepted, laundry, etc.).
3-How can you help them accomplish their goals?
Many ecommerce accommodation type sites such as https://www.marriott.com/default.mi offer filtering of various comparable features while other types of ecommerce sites related to products such as https://www.wayfair.com/ also provide filtering towards desired products. The business benefit of online filtering type features provides the target audience with the ability to more quickly find the desired type product/service that ultimately results in purchases and increased company sales.
Booking.com offers it users recommendations based on a set of user specified “filters” such as location, booking dates, price ranges and so forth (AI Ukraine Conference, 2018; Data Science Summit, 2018).
As with all growth organizations, the Booking.com website continues to enhance it product and service offerings to meet market demand and revenue growth opportunities. Booking.com continues to expand its portfolio of the types of properties that property owners offer to clients seeking to book stays at these properties. Hotels were the first type of properties listed. Competing services such as https://www.airbnb.com/ started with rooms within people’s homes. Market forces and customer convenience forced companies in this space to expand the types of properties offered for booking as a type of one stop shopping for persons looking for accommodations.
The ability for its users to filter a set of booking “features” is imperative as it allows the users to find their desired property more quickly among a set of booking.com recommendations. However, types of properties have their similar and unique sets of features of interest to users. Filter selections for an ever increasing set of available features across property types becomes overwhelming. The need to provide a consistent preferred filter list of features of interest to the user that enhances the user experience.
This business problem of Accommodation Filters is affirmed by the work of Principle Data Scientists at Booking.com (Bernardi, Estevez, Eidis, & Osama, 2020). They state that filter recommendations impose certain consistency constraints related to returned recommendations. The challenge is the Feedback Loop between the UI and their Recommender System. The Feedback Loop is the sequence of:
1-the UI requests recommendations (opens the loop)
2-the System recommends filters based on the query, context, and filter features
3-the UI provides feedback (closes the loop) 4-the System used closed loops to update the model. To resolve this, the system’s architecture was enhanced to implement the Feedback Loops which consists of 2 main components - the Instrumentation Layer used by the UI to integrate recommendations in the platform and Distributed Online Machine Learning responsible for maintaining a model to recommend filters when requested by the Instrumentation Layer and updating as soon is feedback is available.
Experiments were performed to validate the new approach using several Online Controlled Experiments. According to the team of Berbardi et al., they concluded “the system is able to make recommendations that are useful for our customers and superior to the ones by the baseline model (2020).”
One source of information was found in a paper presented at workshop the 14th ACM Conference on Recommender Systems conduction as a Virtual Event held September, 2020 that discusses the Implementation of an Accommodation Filters Recommender System.
References:
AI Ukraine Conference. (2018, November 1). Lucas Bernardi. 100 MACHINE LEARNING MODELS RUNNING LIVE: THE BOOKING.COM APPROACH[Video File]. Retrieved from https://www.youtube.com/watch?v=FtX8jGCTT70
Bernardi, L., Estevez, P., Eidis, M., & Osama, E. (2020). Recommending Accommodation Filters with Online Learning. In 3rd Workshop on Online Recommender Systems and User Modeling (ORSUM 2020), in conjunction with the 14th ACM Conference on Recommender Systems, September 25th, 2020, Virtual Event, Brazil. Retrieved from http://ceur-ws.org/Vol-2715/paper3.pdf
Data Science Summit, (2018, August, 20). 100 Machine Learning Models running live: The Booking.com approach - Lucas Bernardi [Video File]. Retrieved from https://www.youtube.com/watch?v=xxZdFLs6LCg
The key recommendation is to continue to simplify the UI. Having used this site numerous times over the last 5 years, complexity has been both its strength and “Achilles Heel”. The benefits of using filters in search criteria for getting a list of recommendations is tremendously helpful. However, restarting search criteria because of “lost” filters has been time consuming. Abandoning potential purchases and switching to a competitor’s website has sometimes been easier given the number of companies providing online accommodation services.