This document describes how latent Dirichlet allocation (LDA) was used to model endorsement data from Booking.com's Destination Finder project in order to optimize user engagement. LDA was applied to model endorsements from over 10 million real user endorsements as mixtures of latent topics. Some key topics discovered included shopping, museums, and culture/temples. Mapping destinations and users to the topic mixtures allowed for personalized recommendations. While LDA worked well, there were challenges with sparse, ambiguous, and competing endorsements that required further analysis and optimization of the model.