Digital View presents the unique thoughts, ideas and experiences of a Digital Public Service team member. Our first subject in this series of blog posts is Elisa Capecci, a data scientist who works as part of the Business Intelligence team.
Who I am
I’m originally from Italy, but I’m a Wellingtonian now. I came over 9 years ago on a sort of adventure — you know, the other side of the world, so exotic — but when I arrived here, I felt at home. Here at the Department of Internal Affairs (DIA), I feel included and supported. Those are big things for a woman and a foreigner.
I did a PhD in Computer Science, specifically in machine learning. My thesis was in spiking neural networks, which are used in applications for machine learning and artificial intelligence. In machine learning, you use some data to train algorithms. You can learn from structured and unstructured data and find patterns where you don’t see them. You can use supervised or unsupervised learning, discovering what is hidden there in the data. It’s really interesting.
In academia, it often takes time to see the end goal of what you produce. A lot of times, what you’re doing might be used in the future, but that gets frustrating. I wanted to feel like what I was doing could have an impact right now, and I wanted to use my skills and passion to influence changes that can help the population. I think government is the place where I can do that.
What I do
I work in the Business Intelligence (BI) team for the Digital Public Service (DPS) branch. My team is quite new, and our main purpose is to provide actionable insights to support the branch in delivering evidence-based advice. Using data helps the branch set its priorities, inform its strategy and shape its decisions.
At the DPS, our goal and mandate is to achieve the outcomes of the Strategy for a Digital Public Service. One of the focus areas is to explore new ways of working to deliver better service for New Zealanders.
Presently, I’m helping the Marketplace team improve the timeliness of the highly repetitive 2-stage onboarding of prospective information and communication technology (ICT) suppliers. Reviewing the content of submitted supplier documents is a critical step to ensure they are accurate and sent to us under the right Marketplace service definitions.
This repetitive process involves several members of the Marketplace team. I am applying mature text mining and algorithms in natural language processing to automatically read and classify the submitted documents to give a level of independent assurance — this speeds up the manual validation process and, more importantly, allows the team to use their time and skills on more productive activities.
This is only the beginning. I’m excited by the prospect of applying machine learning techniques to improve other Marketplace processes, but also to meet the needs of other business units within DIA.
How machine learning works
The first thing to understand about this is: data is not just numbers. Text mining and natural language processing have been around for decades. We use them every day — how good is text prediction, even when we write an email?
I mine text from the supplier applications, where words are used to describe a service. I input these words into my classifier, so these words are my data. I train my classifier to recognise the data. The more documents it reads, the more it learns, and therefore the better it functions. Every new document that comes is assigned to one of the learnt categories.
People are used to seeing results from data as bar charts and line graphs, but a digital dashboard gives us the chance to deliver clients a tool they can interact with. You can highlight part of it, you can change the parameters — it’s a big advantage over simply printing it out.
At the moment, we’re providing only descriptive analytics, such as summary views, on data we have collected over a certain period. By adding insights from machine learning techniques, we can go deeper, we can optimise our results, we can predict and forecast. Machine learning is powerful because we can personalise the results we populate to dashboards.
I believe that one of the BI team’s main tasks is to influence the growing data culture in the branch. Data can really help us to improve our work, the way we make decisions and the evidence we provide. It can also help us to better plan the work we do.
I know in the branch that people have great ideas and great questions they want answers to. Questions like:
- are we doing a good job
- how are we measuring this
- are we achieving the outcomes of the strategy?
This is where forecasting and prediction can be helpful. It’s not just about looking at what happened in the past, but also what the future could look like and how we can optimise it.