This talk describes the factors affecting travel-site user behavior, how understanding that behavior can improve ranking & recommendation engines, and what the challenges are.
Detailed abstract: In the past few years, many travel search engines have been developed, aiming to provide travelers better experiences. Several factors affect users’ selection for journeys: location information, provider reputation, price, journey availability time, number of stops, and more. Understanding the user behavior can help in two key ways: better ranking for search results to make sure the most convenient journeys appear in the top of search results, and to recommend to users similar journeys that they may be interested in. By collecting information about user behaviors (e..g click through data, explicit feedback, page visit time) along with journeys and providers information, this is a good basis for building ML models for ranking and recommendation. The main challenges for building these models are: converting user implicit feedback (clickthrough data, page visit time) to reliable relevancy score; model selection; measuring model performance; and online learning (developing the model to incrementally improve over time).
Bio: Mohamed Baddar is a Senior Data Scientist/Engineer at GoEuro GmbH. There, he develops recommender systems for travel search engines, using R, Spark, Scala, and Java. Previously he was a Senior Data Engineer at BADR, an Egyptian Big Data Startup, where he developed a recommendation engine for e-commerce, a forecast model for e-commerce. Before that, he was at IBM, doing predictive analytics & ML for drug demand forecasting, and text mining of software defects.