What’s the culture like in our Insight & Decision Science team? Xiaolan reveals all.
“I really enjoy the working life here. It’s a place full of people with curiosity, passion and innovative ideas”
Name: Xiaolan Sha
Job Title: Lead Data Scientist
What area do you work in? Data Science team - Insight and Decision Science. Our team is involved in a variety of different activities. In the last month, I’ve personally focused on building models to support content analytics and marketing. The core idea consists in figuring out new and exciting ways to understand customer behaviours, and leverage the collected insights to broaden the content reach.
What's your background? Where did you work before and how does it differ to here? Similarly to many of my colleagues, before joining, I followed an academic course. I have a PhD in computer science, and I worked in a research team focused on designing context-aware recommender systems. There are many differences of course between working here and pursuing an academic career, yet there are some similarities. I try to always keep up with the research community and follow the state of the art. One of the main challenges in doing research in these fields is having access to relevant and significantly large corpuses of data allowing to make meaningful observations. This is never a problem here: we focus on the challenges associated with having to handle immense data volumes on a daily basis. It’s amazing to deploy your own models into production for business applications, and it’s also very exciting to measure the impact of your models on real customer interactions.
Tell us a bit about your most recent project? Content is core to us. We have a collection of carefully selected high quality entertainment, movies and sports content. It’s not possible to present our entire media selection in a single UI screen to our customer, so we need to help the users to discover content that may be of interest to them. My most recent project is to help the marketing team to broaden the reach of the content and help users make new discoveries in our media selection. We do this by building a recommender system with an ensemble of algorithms that generates personalized recommendations in different contexts and scenarios.
Why was the experience on this particular project worthwhile? At first glance, the project may look like a traditional and “simple” content recommendation problem, however once you get into the details you realise you’re facing a number of new challenges on a variety of levels. At algorithmic level, approaches exist to recommend existing content to existing users, but what about the release of a new title? How can I decide which users across my customer base are more likely to appreciate a new, upcoming TV show? Also, at implementation level, a lot of problems that are usually treated as trivial in textbook examples become tricky to handle at scale. We have sometimes hit system limitations, and we have had to optimize the scalability of our machine learning algorithms built on top of Apache Spark. However, after facing all these challenges and all the frustrations associated to these “real world” problems, seeing an actual customer uplift as a result of your work is a great satisfaction! The models we have built have been regularly used by the marketing team to carry out personalized campaigns, and they could be now automated and extended to supply the OTT platform once it will be deployed in production.
What did you enjoy most about working in IDS big data team? There are three main things: First and foremost, I’m constantly exposed to new technologies and big data tools like Apache Spark for machine learning and streaming, and potentially Tensorflow for POCs. Secondly, the team culture is great: we are a team of like-minded people who are passionate about modelling real-world problems with mathematical formulas, and we always keep learning to improve our expertise. The third would be the working life here. It’s a place full of people with curiosity, passion and innovative ideas. For me, the fun part is all about delving into the vast amounts of data from all over the business to search for an answer.