AI and machine learning are two of the hottest buzzwords around, but for our Data Scientists they’re much more than that. We talked with 4 of them to learn more about the joy of unearthing hidden treasures through data
Imagine facing an enormous, ever-growing mountain of straw and being asked to extract the valuable gold threads that are hidden deep inside of it. How do you approach such a task? Thoughtfully is probably one of the first words that come to mind. When dealing with such vast amounts of data – every day we obtain over 3TB of it – you don’t just dig and hope to get lucky. What you need is a machine that knows how to sort gold from straw. Unfortunately, since these types of machines aren’t exactly an over-the-counter item, you need to build them from scratch. And that’s what our Data Scientists do.
Hila Gat, head of Data Science at Wix, says that the group grew exponentially in recent years. “In the beginning we worked on well known AI problems that had to be fine tuned to our needs. For example, In the area of computer vision we created models for image segmentation, super-resolution, and automatic cropping. We built and tailored the solutions to the Wix product, along with achieving better performance than was offered off the shelf. We also had to put a lot of effort in collecting quality data where it was not available. In order to enhance images, for example, we have a group of designers whose sole purpose is to photoshop images in order to ‘teach’ our model what an enhanced image is, and how to do it by itself. Today, our projects are a combination of long-term research problems that were not solved yet by any company, and shorter term products solutions”.
On top of that, we work on complex AI search capabilities and on Natural Language Processing (NLP), which enables our chat bots to recommend a relevant solution to a user who’s in need for help. This type of ability enables our users to get an immediate solution to whatever’s bothering them while also taking off load from our Customer Care Experts”.
Connecting machine learning to the business environment
Gilad Barkan, one of the group’s team leaders, has been involved in machine learning, image processing and computer vision for more than 23 years. After a brief startup experience in which he decided to dive deeper into the brave new world of AI, he realized that although he already had a Bsc in computer science, he had to go back to school in order to become an expert in the emerging new field. He got an MSc in computer science and applied mathematics and became interested in applying machine learning in the business environment.
Gilad joined Wix about 2.5 years ago when the Data Science group had less than 10 Data Scientists. Now they’re over 60 scientists, Data Labelers, Data Curators and engineers, and still growing. “While some companies use data science to solve one big problem, at Wix we have many of them so one of the main challenges is giving each one enough attention”, Gilad laughs. “We want Wix to think AI and understand what can be achieved through it. I’ll give you few examples: one of the things we can do is to ‘read’ the content on Wix’s hundreds of millions of websites and to categorize them into groups and identify emerging trends; Another aspect of it is computer vision: many of our users have pictures on their sites, but many of those pictures are pretty low quality and we can fix them through AI.
These days, everybody’s saying that data is gold, but in its raw form it’s very hard to digest. What we need to do is locate the meaningful data and to turn it into insights we can act upon”.
What kind of people are you looking for?
We’re looking for humble, fun-to-work-with people with the potential of becoming unique at what they’re doing”, Gilad says. “Data Scientists are the people who phrase the questions and create tools that will enable them to eventually get the answers. They should be the best statisticians among software developers and the best software developers among statisticians, while also being able to talk to people on the business side of things and to communicate their results to different parts of the company. We’re committed to give them whatever’s needed to turn them into an end-to-end Data Scientist”.
Data by itself isn’t enough
Noa Stanger, the group’s operations lead, says she found herself in the team almost accidentally. “About 5 years ago I was an architecture student and had a job as a Customer Care Expert at Wix. Later on, after I became a shift manager, I understood that I really prefer the Wix environment and its pace compared to the architecture world. After I graduated, I started looking around the company for another role, and then got an offer to join the Data Science team. We were only 14 team members back then – all of them developers and Data Scientists – and we understood that we didn’t have enough available data. That meant that it had to be gathered and maintained, and so I became the team’s first Data Curator and then I built the data labeling and curation team.
Learning machines are supposed to emulate a human’s decision-making process, but if the data we feed into those machines isn’t good enough, even the best mathematicians in the world won’t be able to help them”, Noa explains.
In order to do that, Data Curators organize and integrate data collected from various sources and Data Labelers add one or more meaningful and informative labels to each file in order to provide context (for example, a “cat” label to an image of a cat), so that a machine learning model can learn from it.
Building such teams, Noa says, is unique to Wix. “You always need to know how to ask the right questions, and after the Data Scientists build a model, you need to figure out a way to test it and see if it works, but we understood that having designated teams to do these tasks is much more efficient for our Data Scientists and produces better results”, she says.
Curious people with a passion to impact our users’ experience
Doron Bar Tov also didn’t start his career as a Data Scientist. He has an MSc in physics, but after working as a physicist for a while, had a change of heart. “I felt that it wasn’t for me so me and a few friends founded a startup company”, he recalls. “After that I transitioned to deal with data and worked for an Internet marketing company and then started at Wix about 5 years ago. The funny thing is that out of the 5 guys who founded that startup, 3 of us now work at Wix, and we are still trying to recruit the fourth one”.
Doron was tasked with finding ways to make financial predictions and about 3 years ago became a team leader. He’s still in charge of that but he and his team members (there are 10 of them), also try to predict some of our users’ behaviors based on data.
Predicting last year’s numbers must have been a bit of a challenge.
“2020 was insane. The pandemic broke all of our models and we had to rebuild everything from scratch”, Doron sighs.
But perhaps because of the unpredictable nature of reality, Doron thinks that the Data Science group at Wix is one of the most interesting groups in the company. “Wix is a very broad product. There are many products within it, and a lot of data is collected. This means that we have a huge potential to create many data-based products on our platform. Anyone joining us will have the ability to influence the way AI products will be built in a very large company, impacting many millions of users. That’s why we’re looking for curious people with a passion to impact our users’ experience and to really move the meter. Another great thing is that apart from professional development days, our team members usually transition from project to project every few months and that way can not only get to know many parts of Wix, but also familiarize themselves with new domains in machine learning. It’s very cool”.
Opmerkingen