Author Spotlight
In the Author Spotlight series, TDS Editors chat with members of our community about their career path in data science, their writing, and their sources of inspiration. Today, we’re thrilled to present our conversation with Lowri Williams.
Lowri Williams is a Wales-based research data scientist who has accumulated four years of experience in academic research and two years in industry-based collaborations supporting small-to-mid-sized businesses with novel data science solutions. She had collaborated with multidisciplinary experts to present research outputs and findings within several reputable academic journals, and has also turned to Medium and Towards Data Science to showcase some novel data science approaches in tutorial-styled blog posts.
What first inspired you to learn data science?
I’ve always been curious about data. I remember when I was 13 years old, as part of the geography curriculum, we had to produce a project on a geography-related topic that interested us. I decided to measure some attributes of a river near my childhood home.
I found a map and took my bike to specific areas of the river and measured how fast the water flow was, how wide it was, etc. And as boring as it sounds, I enjoyed it. I spent many hours hypothesizing and mapping and graphing the data to demonstrate my results. I handed in my report with pride and received the highest mark in the class. I was proud. Fast-forward 17 years later and I’m a research data scientist.
Can you share more about the path that led you to where you are currently?
My route to becoming a data scientist was quite a simple one. I spent three years studying for a B.Sc. in Information Systems at Cardiff University, Wales, UK. The core of the degree was about how humans interact with technology and information. It was less technical than I hoped it would be and what I could do. I wanted something more challenging, something to force me to learn and take responsibility for my own growth. So… I applied for a Ph.D. scholarship and ended up getting it!
I have to admit, I wasn’t really prepared or didn’t quite know what I was getting myself into. I had some rough idea as to what I wanted to focus my research on, but how to research and do it properly was a huge learning curve for me. But, sometimes diving into the unknown is the best way to not only learn about a topic but also to learn about yourself.
My research interests revolve around natural language processing, in particular sentiment analysis. I guess this interest goes hand in hand with my bachelor’s degree – there’s something about understanding how humans communicate with technology that intrigues me, in particular the language they use.
Once I had graduated with my doctorate, it was time to jump into the daunting pool of job hunting.
Many early-career data scientists are worried about landing their first job—what was the experience like for you?
I found it incredibly hard to place myself in a role that I found interesting or that I felt like I could do it justice. I wasn’t just someone who knew about natural language processing, I was more than that. I was someone who knew about machine learning, feature engineering, classification, annotating data, how to generate strong datasets to achieve stable machine learning models, how to interpret and visualize these results. I knew how to clean and process data, not just textual data. I was a coder, an implementer, a developer, a problem solver.
I was all of these things in my own way, that classifying myself as a software developer, a data engineer, a this, a that, was really, well, hard! I spent a few fair hours trying to define who exactly I was and I am today, and that is a data scientist. And I think that’s ok – we all have our own timelines when it comes to these things. I know people who knew from the very first day of their bachelor’s degree what their next steps in life were going to be, and they’ve followed through with them. I also know others who change roles every two or three years because they often need more stimulus or something more challenging for them to help them grow. Either route is ok, and I think this is something I would tell my younger self if I was to ever time travel back to that time.
So where did you end up?
I’m currently a research data scientist as part of a project at Cardiff University that supports industry based small-medium businesses with novel data science solutions. And I like it! I like being able to try different ideas and approaches to different problems – not to say it’s a playground of data or anything, I have produced some projects which have helped improve businesses! But I have room to be creative with data – something which I’ve learnt to be quite an important skill as a data scientist.
At what point did you decide to start writing publicly, and what drove you to make that decision?
It was a little daunting at first to turn to Medium and Towards Data Science as a writing platform. As an academic, it’s almost expected that we publish scientific literature which follows quite a crude format of "here’s our hypothesis, here’s what others have done, we don’t think this answers our hypothesis, therefore here’s our new approach, here’s the experiments, the results, and knowing this now, what we might do in the future if we ever get the chance in between our heavy workloads to do it…"
Because of the opportunity to be creative within my current role, I felt like this didn’t quite fit in the academic world. Not that my approaches to problems are wrong or anything, because they solve the problems. But they’re less experimental in the sense that they follow more of an agile approach where we face one problem, solve it, and move on to the next based on the outputs of the other solutions. Defending some of the reasons from an academic point of view therefore becomes more challenging.
So, I turned to Medium to share these outputs and some of the methods that I’ve explored. It also allows me to practice a different writing style than the one I’m used to. I wouldn’t say it’s easier to write a blog post either. To write clearly and consistently is sometimes challenging! But I like the challenge as it’s opened up collaborations with other contributors with whom I share ideas and converse with.
What kinds of changes do you hope to see in the data science world in the next few months or years?
I really hope there’s a shift in the natural language processing field soon as it’s been relatively similar for a few years now. And it’s coming. Word embeddings and BERT are stomping their way through and showing us how deep learning models can produce some really cool results. I think this is the general case in machine learning. For example, there’s an abundance of machine learning approaches that have been proposed to help automatically triage COVID-19 patients from the sound of their coughs, or to automatically classify whether a patient has pneumonia based on their chest X-rays. So, I’m looking forward to getting my hands dirty with similar models and seeing how we can use them for the greater good.
If you’d like to learn more about Lowri’s work and the way she leverages her academic expertise to approach real-world challenges, here’s a selection of her TDS posts from recent months. You can find the full archive on Lowri’s profile, and links to her academic publications on her website.
- Sentiment Analysis: Idioms and their Importance (TDS, September 2020)__As good an introduction as any to Lowri’s research and the ways she translates it into engaging, accessible posts, this article is based on Lowri’s doctoral research. Focusing on idioms’ role and value to sentiment analysis, it’s a stepping stone towards gaining clearer insight into the way people use everyday language.
- Spotify Sentiment Analysis (TDS, May 2020)__Part fun side project, part hands-on tutorial, in this post Lowri turned to her own Spotify data to try to detect patterns and insights from her listening habits. (Spoiler alert: it turned out Lowri’s favorite music is happier than she’d expected!)
-
[WordNet](https://wordnet.princeton.edu/): A Lexical Taxonomy of English Words (TDS, October 2020) Going back to NLP, this post is a thorough explainer on WordNet, Princeton’s lexical database. Being a Lowri article, it doesn’t stop at high-level theory—prepare to roll up your sleeves and experiment with some useful code examples.
Stay tuned for our next featured author, coming soon! (If you have suggestions for people you’d like to see in this space, drop us a note.)
