aDFR: An Attention-Based Deep Learning Model for Flight Ranking

Although there are numerous available flights for travelers on the online ticket reservation platforms after entering a query, they often make decisions only based on the information of a few flights that are ranked high on the list. Therefore, it has become an important strategy for online travel agencies to rank flights according to their potential popularity. However, to build a ranking function for flights is not an easy task because the preferences of passengers vary from flight routes and change with time. In addition, the selection of a flight is affected by the information of other flights in the same returned list. Traditional Learning to Rank (L2R) methods fail to model the distinct preferences of passengers adaptively on each query. They are also insufficient to capture the complex relationships among flights in the same list. To cope with these challenges, we design an attention-based deep neural network for flight ranking. It adopts two kinds of attention mechanisms to model the dynamic preferences of passengers and the relevance between flights respectively. Extensive experiments on real-world flight order datasets demonstrate the superiority of our method against other competitive ones.