There are no aliens working in artificial intelligence (AI). We repeat: no aliens.But we do have Machine Learning Engineer Rosa Maria Siervo. She uses AI to serve our readers the best possible recommendations. Beam us up, Rosa!
To bring some structure to DPG Media’s IT organization, the organization is split into five clusters, each with several areas. So within the Digital Enablers cluster, we have the Recommendation & Search Area. It’s a relatively new area, but one with lots of history. In January 2022, Dutch and Belgian teams already working with recommendation models joined forces. Rosa is happy with the new set-up: “We learn a lot from each other through regular meetings and spontaneous conversations.”
The area has several squads with a specific focus, like video recommendations or news personalization. NPS (News Personalization Squad) is Rosa’s squad. Simply put: they try to understand the content of news articles and a person’s reading preferences to give the reader the best possible recommendations when they navigate the app or website.
Rosa: “The work we do is very similar to my colleagues' work on video recommendations: we both work with artificial intelligence. But there are a few differences. An important role of news recommendation is to optimally balance between providing news that interests readers, news that provides alternate views or opinions, and news editors believe readers should know about. Another difference is that with video, you can recommend old videos too, but with news, you want freshness.”
Rosa explains the challenges of fresh content: “In video recommendation, it makes sense to recommend videos that users with a similar watch history also interacted with. In machine learning terms: collaborative filtering. But in news, we have the item cold-start problem: when a news article gets published, we have no interaction data on it yet. Nobody has read the article. That's why we employ a diverse range of algorithms to tackle typical situations in the news domain. In the absence of interaction data, we rely on content-based algorithms, which focus on the article's content and the user's individual behavior. When sufficient interaction data is available, we strategically leverage either collaborative filtering algorithms or content-based algorithms based on the specific use case.”
Rosa’s squad started with NU.nl and then expanded to HLN.be, AD and ADR brands. The aim is to incorporate recommendations on all of DPG Media’s news brands and also some magazines."“One at a time,” Rosa smiles. It’s not like you can copy-paste an algorithm to a new platform. Actually, it’s even more complex. Rosa: “Our algorithms can be excellent right now, but people’s expectancies always rise. For example, in e-commerce, product recommendations are very accurate, so people demand that same level from us. So we keep monitoring and improving our systems based on new data and research.”
It’s not our job to create filter bubbles; it’s our job to prevent them.
Fighting filter bubbles
Much has been written and said about filter bubbles and how they can distort people’s sense of reality. The squad’s commitment is to inform people in the best possible way - no filter bubbles, no misinformation. “It’s not our job to create filter bubbles; it’s our job to prevent them. We want to maximize user engagement and make sure people are well-informed. So, we do recommend some articles to add diversity to the recommendations. It’s a complicated topic, and we continue to research it.”
The squad runs a program with the University of Utrecht to determine how to measure diversity and define systems that balance accuracy and engagement with diversity metrics. Rosa enjoys the collaboration a lot and sees it as a win-win for both parties: “they are up-to-date about current research and share that knowledge with us, and we provide the real-life data for their research.” And let’s face it, it’s a busy job, so getting some extra hands and input is always very welcome.
A perfect match
“My job is a perfect mix between science and engineering. I train and build models, deploy them, ensure low latency, and more. I get to work with many technologies and on the architecture too; the big scale makes it challenging. But I still get that research side too. Not only through collaborations with universities, but also by talking to industry peers at conferences like RecSys (Recommender Systems).”
It’s never a 50/50 division between science and engineering. Rosa: “Sometimes, I’m really into the research part. Like right now, I lead the research program with the university and mentor our interns on their research topics. So now I read more papers than usual - I even find reading papers relaxing! But not long ago, I focused on getting my Associate Solutions Architect certification from AWS.” Mission accomplished, by the way.
For the love of Dutch
The balance between engineering and research is what convinced Rosa to join DPG Media. And the Dutch language. Italian-born Rosa is into natural language, especially the Dutch language. “I am fascinated about understanding hidden patterns within a language and how I can make a model understand those patterns from text.” Sounds like a match made in heaven.
Rosa saves the Dutch language for her models as her team’s go-to language is English. She has a very international team: Korea, South Africa, Romania, India, the United States, the Netherlands, and Belgium - they make for a good mix. “I feel 100% part of the company, and language is not a problem. Some e-mail communications and systems are in Dutch, but if it’s important, colleagues translate the main takeaways and put it in our Slack group. If someone starts a meeting in Dutch, a Dutch colleague has already asked to switch to English before you know it.”
Confidence levels through the roof
Before DPG Media, Rosa gained work experience in machine learning at other Dutch companies. But she never experienced the opportunity to grow as much and as fast anywhere else. “I’ve always been into the tech side of things, but I was insecure about my technical skills. I received so much support from everyone here; it made me more confident. There’s no such thing as a stupid question within my team. I can always ask for help, and you don’t realize how fast you improve your skills by getting immediate feedback. In the past, I was afraid of making mistakes. Now I finally understand the concept of failing fast instead of not failing at all.”
Her newfound confidence helps Rosa give back to communities like Girls in ICT and New Dutch Connections. She tries to inspire others to start coding and excite women about working in tech. “Don’t be scared of working in AI - we’re not aliens. If you like technology, you can definitely do it. All you need is some programming experience in, for example, Python. Put in the effort, and the rest is passion. I think that’s all it takes to start a career as a Machine Learning Engineer.”