Optimization of a flight recommendation system

Abstract

When a Travel Website requests for travel recommendations to go from A to B on a given date, there are potentially thousands of possible itineraries. Simply returning the 100 most unexpensive flights is not the most relevant answer: people also make their choice based on their convenience (E.g., How long? How many connections? What is the departure time?). The problem herein is not to select the best flight, but an optimal subset that has a high probability of containing that flight. Search engines and recommendation systems are already implemented in the market e.g., Netflix, Google, Amazon, etc. Therefore, a challenge is to apply this approach to the travel industry. The Search, Shopping & Pricing (SSP) product engineering group serves the core Distribution business of Amadeus as well as Airline IT e-commerce business and the Rail IT and Travel Intelligence new businesses. It is responsible for the travel Search and Shopping products, which power today a vast number of airline and travel agency websites. Based on past bookings, the main goal of the work presented in this final paper is to critically analyze one of SSP’s core products with the aim of enhancing their algorithms and optimizing Amadeus’ search engine to better return travel recommendations. The used methodology is not just to rank the recommendations, but to find the subset of recommendations that is the most likely to lead to a booking. It is introduced the concept of a value search engine and show how this system is based on reservation probability. Then, it is adjusted some of its computing parameters, run different combinations, compared their results, and selected the most effective. At the end of the paper, I aim to obtain considerable results that show the growth of the algorithm's performance, making the concepts of subset selection simple and understandable to implement in a recommendation system.