Keeping track of the latest news on social media is daunting enough, so imagine if your job were to dissect and analyze every single post, tweet and Instagram story on the internet.
Enter the data science and analytics team at Networked Insights. The team develops technology to make sense of billions of daily posts, categorizing the text using more than 30,000 classifications to help companies better understand their customers.
Four members of the company’s data team gave us the lowdown on how they extract insights from social media — and the machine learning tools they use to get the job done.
EMPLOYEES: 73 (61 local)
WHAT THEY DO: Networked Insights provides social media analytics to help marketers better understand their customers, inform their media spending, invent new products and drive creative campaigns.
WHERE THEY DO IT: Chicago
NEW PARTNERS: American Family Insurance acquired Networked Insights in 2017.
THE PERKS: Employees receive unlimited time off and free breakfast on Fridays.
Bryce Lobdell, Lead Data Science and Analytics Group
Bryce Lobdell leads Networked Insights’ data science and analytics team, managing projects, sanctioning proofs of concept, providing technical assistance and architecting systems.
BEYOND WORK: Bryce plays in sports leagues around Chicago, including in basketball, flag football and ultimate frisbee.
What problems are you currently working to address?
Our technology helps customers understand what is being discussed on social media. We collect as much text from the internet as we can ethically and legally buy, and then classify the person who wrote the text and its content.
We have about 30,000 text classifiers, which compose our taxonomy — a hierarchical list of concepts, topics and attributes. At the center of problems we’re solving is how to make smarter text classifiers that are both highly selective and sensitive. We have several types of text classifiers. Some are simple and work for easy cases, others are powerful and complex.
At the center of problems we’re solving is how to make smarter text classifiers that are both highly selective and sensitive.”
How do you prepare your team to tackle these huge data problems?
Machine learning and data science require a wide array of skills and experiences. Our goal is to have complete, redundant coverage of those skill sets. This is too expensive to accomplish solely through hiring, and too difficult to do at our scale. So we set aside time to make sure multiple employees get experience with a particular practice or toolset.
We’re also developing a literature diet to stay up to date on the most productive tools and techniques. Three team members recently attended a computational linguistics conference and brought back several techniques we plan to explore in a proof of concept. We are always looking to improve our training through new learning materials and a high-quality set of code examples.
What accomplishment makes you the proudest?
We developed a tool that gave us new capabilities for language classification. The tool gives us better selectivity and the ability to extract entities from sentences. I was very pleased with how many people on the team contributed to the project. The workflow was close to what I view as a professional workflow for data science and neuro-linguistic programming work.
Drew Boshardy, Machine Learning Engineer
Drew Boshardy focuses on the implementation of machine learning models, which includes developing the supporting infrastructure and data that drive them. His team works to classify social media posts based on the topics or interests mentioned in them.
BEYOND WORK: Drew sings in the Chicago Chorale and helps coach the group on Russian language pronunciation.
How has Networked Insights helped you grow in your career?
I started working at Networked Insights five years ago as a data analyst. I finished my master’s degree in computer science while working here and then came to a crossroads. I wanted to stay at the company, but I also wanted to be a software engineer. I spoke to our CTO, Brad Burke, and he had a lot of experience and advice to share. Ultimately, I was able to stay with the company as a software engineer.
I was lucky that enough people in the right places saw the potential in me and helped me through that transition. I love working at Networked Insights because they value that entrepreneurialism and collaboration. I enjoy trying out a new idea, showing it to people and convincing them of its merits. I’ve been able to learn so much more than I would if I had to start fresh somewhere else.
What’s the most challenging project you’ve tackled?
My biggest challenge was building our internal reporting tool for client success. I was able to build off our existing architectures, but the rest I had to create from scratch.
Writing the various jobs for each report formed the bulk of the work. I started with Apache Spark, but once I was introduced to Apache Beam, I realized it was much better suited to the reports — both in terms of complexity and cost. This project was also one of the first times I used Kubernetes, so that added to the complexity once I got the system running.
We’ve managed to develop a linguistic rules engine that can scale to billions of documents.”
What accomplishment means the most to you?
We’ve managed to develop a linguistic rules engine that can scale to billions of documents. We also built it so our experiments are repeatable. We ensured that every trial we run with the rules engine can be tied back to the code and data that was used to get those results.
Josh Oberman, Machine Learning Scientist
Josh Oberman is responsible for building new machine learning classifiers, as well as monitoring and updating the old models in Networked Insights’ code base.
BEYOND WORK: Josh enjoys playing guitar and going to live shows.
We understand you’re newer to the team. What brought you to Networked Insights?
I have an academic background in philosophy of science and artificial intelligence, and have worked as a data science and machine learning practitioner in the industry. I joined Networked Insights because of the people who work here and the company’s use of cutting-edge technology. The onboarding process was smooth and casual. All new hires are tasked with going around with a drink cart on their first Thursday. It was a great icebreaker.
What do you like most about your job?
I spend my days either researching, programming or debugging code. Coming from a research background, I love that I have access to a lot of tools and the leeway to use cutting-edge methods.
I’m currently building a universal language model deep neural net in PyTorch that I will use for transfer learning on a text-based classification problem. This method is based on a paper that came out this past April, and I have all of the resources of Google Cloud Platform at my fingertips to prototype the model and implement it at scale.
Some companies preach giving engineers ownership outside of relatively narrow technical problem domains, but few truly practice it.”
What values does the data team hold?
Individual ownership and responsibility. I set my own hours and have a lot of flexibility to set my own agenda relative to the broader business goals. Some companies preach giving engineers ownership outside of relatively narrow technical problem domains, but few truly practice it.
With that said, collaboration and staying in active conversation with other colleagues is just as important and necessary as owning your own work. I’m lucky to have colleagues who are intellectually curious and willing to lend a helping hand. If there are any markers of success for someone joining this team, it’s being solutions-oriented, open-minded and collaborative
Majed Takieddine, Data Analytics Engineer
Majed Takieddine works on a variety of projects that range from machine learning to software engineering, data analysis and visualization. The common thread is that they all help Networked Insights process data.
BEYOND WORK: Majed is an avid traveler, having visited 35 states and 18 countries. He enjoys experiencing different cultures and meeting new people.
What’s your tech stack? How does it help you approach your work and innovate?
Our focus is on distributed learning, large-scale model building in machine learning, artificial intelligence and natural language processing. To accomplish this goal, we’ll use systems such as Apache Beam DataFlow, BigQuery, BigTable and Kubernetes for big data.
We’re also proficient in Python, SQL, Java and Groovy, with extensive experience in Scala. We gain hands-on experience in building and deploying models through the latest advances in technology, including deep learning and reinforcement learning. I’m always learning the newest technologies and tools through books and online courses. Without the constant self-driven learning, I wouldn’t be able to innovate and be successful at my work.
Without the constant self-driven learning, I wouldn’t be able to innovate and be successful at my work.”
What mentorship opportunities are there at Networked Insights?
This summer, we had two interns working with our data science and analytics team. I’ve been their mentor, and it has been a rich learning experience for me. Through online resources and sitting next to them to answer any questions or address problems they encounter, we helped them build a full working model that can classify toxic, profane and bullying social media posts.
If you look back at what your team has accomplished, what makes you the proudest?
I’m proud of everything we do, but what stands out the most to me is the strong team spirit, motivation and easy-going communication between team members. When I started here, the data science and analytics team didn’t exist. We were split into small groups under different managers, so we didn’t have a strong team spirit. Since we formed our team, work productivity has increased and we’ve become closer.