We introduce a novel dataset of real multi-destination trips booked through Booking.com's online travel platform. The dataset consists of 1.5 million reservations representing 359,000 unique journeys made across 39,000 destinations. As such, the data is particularly well suited to model sequential recommendation and retrieval problems in a high cardinality target space. To preserve user privacy and protect business-sensitive statistics, the data is fully anonymized, sampled and limited to five user origin markets. Even so, the dataset is representative of the general travel purchase behavior and therefore presents a uniquely valuable resource for Machine Learning and information retrieval researchers. This work provides an overview of the dataset. It reports several benchmark results for relevant recommendation problems obtained as part of the recently held Booking.com data challenge during the WSDM WebTour workshop.

Image galleries provide a rich source of diverse information about a product which can be leveraged across many recommendation and retrieval applications. We study the problem of building a universal image gallery encoder through multi-task learning (MTL) approach and demonstrate that it is indeed a practical way to achieve generalizability of learned representations to new downstream tasks. Additionally, we analyze the relative predictive performance of MTL-trained solutions against optimal and substantially more expensive solutions, and find signals that MTL can be a useful mechanism to address sparsity in low-resource binary tasks.