TripAdvisor Data Scientists Present to CDS Students
TripAdvisor’s presenters focused on machine learning for experiences and rentals on their site
Two data scientists from TripAdvisor, the world’s largest travel website, flew in from their Boston office to speak to a full audience of CDS students. Cyprien de Lichy, Senior Data Scientist, and Peimeng Sui, Data Scientist and 2017 CDS graduate, discussed their roles and experiences at TripAdvisor. De Lichy gave a broad overview of their data science projects which include recommendations, ranking algorithms, search engine marketing bidding, fraud detection with natural language processing, machine learning for optimizing users’ search experiences, and computer vision projects.
De Lichy explained in detail one project involving the sort order of shelves (a term which refers to the website display, similar to Netflix’s shelves for shows). TripAdvisor has over one hundred shelves for each destination — the problem is uncertainty about how the shelves will perform. De Lichy and his team sought a solution with an in-context multi-armed bandit approach which seeks to maximize the sum of rewards over time. The ultimate goal of this approach is to display the optimal shelves while also displaying new, untested content. This requires modeling shelves’ click-through-rates and regular re-training.
Sui discussed his experience of being hired for a full-time position at TripAdvisor after working as an intern last summer. His work now focuses on the recommendation and personalization capabilities of TripAdvisor’s data science platform. Recommendation engines are supported by machine learning models which have scheduled weekly training based on users’ search history on the website. This determines content which can be displayed on multiple areas of the site and in email campaigns.
The overall data science platform, according to Sui, incorporates model deployment, code optimization, and monitoring, scheduling, and serving models in real time. “I use what I learned at CDS every day at TripAdvisor,” said Sui. He added that his favorite aspects of his job are the diversity of machine learning projects, the strong support from engineering teams, the personal growth opportunities, being able to see his work impact millions of users on the website, and company perks. The data science workflow he laid out for prospective hires begins with research and continues to data exploration, modeling, productization, and A/B testing for result analysis.
By Paul Oliver
