Predicting User Flight Preferences in an Airline E-Shop

摘要

With the continuous development of the Web, it is becoming increasingly important for e-shops to present customers with the most relevant products. The personalisation of product rankings is one of the problems associated with this development. Until now, the focus in the field of machine learning has mainly been on the ranking of documents The ranking of items in general asks for new types of features, that accurately describe the match between query and item. We propose the usage of cross-terms between item-specific and user-specific variables in the Ranking SVM algorithm. We apply these new features for the ranking of flights on the website of a company in the airline industry. For our data, the cross-terms improve the out-of-sample accuracy of the Ranking SVM with 2.11% points compared to a baseline. Due to the high amount of traffic on the Web, improvements like this can already have a big impact on users' purchase activity.