ICLR 2019 Expo Talk

-

Expo Schedule »

Computational Problems in Travel Marketing

Sponsor: Expedia, Inc.

Organizers:
Roopesh Ranjan (Expedia Inc.), Angelo Restificar (Expedia Inc.)

Presenters:
Roopesh Ranjan (Expedia Inc.), Angelo Restificar (Expedia Inc.), Eric Lambert (Expedia Inc.), Samira Daruki (Expedia Inc.)

Abstract:

As one of the world’s leading brands in online travel, Expedia deals with data in the Exabyte scale. Expedia serves over millions of customers across the globe with over 350 million stayed nights and over a few hundred million transactions in different modes of travel like flights, car rentals, cruises, trains, activities and more. The solutions we design to solve some of the most complex computational problems involve a variety of machine learning sub-areas including supervised/unsupervised learning, reinforcement learning, deep learning as well as large scale optimization techniques.

The first area of this talk will focus on problems such as click-through-rate and conversion-rate prediction models, and techniques to determine optimal bids in online advertising auctions. We will discuss approaches to solving these problems using gradient boosting, neural nets and time-series forecasting methods. We will present some known problems such as data sparsity, seasonality in demand and dealing with non-stationary auction dynamics and lead discussions on approaches to solving them.

In the next part of the talk, we will discuss personalization challenges in reaching to customers in effective ways to improve customer experience and surface relevant content. We will brainstorm some key challenges in adopting various approaches in recommender systems, ranking, and contextual bandits.

Lastly, content is extremely powerful in how users pick travel destinations and make choices about the modes of travel. We use a diverse set of data from hotel images, reviews, ratings, and other categorical data to build image understanding, sentiment analysis and query understanding methods. These vision and NLP methods that we build have helped us conceptualize and design products such as chatbots to answer travel queries, travel summarization and automated travel planning.