Automated Topic and Trend Detection in Customer Comments Using Natural Language Processing (NLP) A Data Science Project by Dominik Finzer

Shortcuts:
Introduction | The Project | Results & Findings | Programme Information | Contact | Info-Events | Applied Data Science Professional Portraits

Dominik Finzer, graduate of the MSc in Applied Information and Data Science at the HSLU, Data Scientist in the Customer Experience & Insights department at SBB 

First of all, tell us something about yourself: What hashtags best describe you?
#StayCurious
#LifeLongLearning
#WhySoSerious?  

Tell us more about the hashtags.
Since I was four years old, my dream has been to become a data scientist (end of irony). To find out if something is for me, I have to try it. As a child, I didn’t know which career was right for me, so I tried a lot of things. Looking back, I have never regretted trying new things and moving on when the time was right. Each job has taught me something and given me something that I wouldn’t want to have missed. Staying curious, getting to know new people and problems, and getting out of my comfort zone has helped me personally to find joy in my work and in my free time. Last but not least, don’t take yourself too seriously. Humour is also very important to me, very important in everyday working life. If you can laugh together, you can face challenges together.

What do you do at SBB?
I currently work as a data scientist in the Customer Experience & Insights department. I work in an interdisciplinary team of data scientists, statisticians, designers and psychologists. Our goal is to bring the customer’s perspective into the company and help to ensure that the wishes and needs of customers are more strongly integrated into decisions and the related services. In my role, I am primarily responsible for providing this customer perspective in the form of data and analysis. In addition to linking data from various customer satisfaction measurements with other company data, a key task is to record relevant questions from the company. With the data available and the issues identified, I help to create a reliable basis for making more customer-centric decisions.

What did you do before and why did you join SBB?
After a basic technical apprenticeship in the automotive industry, I studied industrial engineering. I then worked as a business analyst and project manager, first at a start-up in Basel and then at an IT consulting firm in Bern. I’ve been with SBB since 2018, where I first worked as a corporate architect. In this role, I learnt that complex issues can be communicated much better with well-designed visualisations. To improve my skills in this area, I undertook further training in data visualisation and met a designer along the way. Together, we decided to take on the position of “Project Manager Data Visualisation and Storytelling” in 2021 as part of a job share. At the same time, I started my Master’s degree at HSLU to deepen my knowledge in the field of data science and, in particular, to improve my programming and engineering skills.

Tell us about your research project.
When measuring customer satisfaction at SBB, in addition to the structured data that comes from customer ratings, we collect a lot of free text comments in which customers express additional concerns. The effort required to write such a comment always costs respondents time. So the hypothesis was that these comments contain issues that are particularly emotional for customers. As these comments are still only read sporadically, the idea was to use today’s natural language processing (NLP) methods to mine this “treasure trove” of data and gain new insights into customer concerns.

For an interactive version of the graphic, please click here.
Topic clusters meet keywords: The graph shows the themes extracted from the comments and the keywords behind them to illustrate which aspects move customers emotionally.

Topic Modeling helps to extract patterns and key themes from large collections of text data. The insights provide a deeper understanding of customer opinions and needs.

What data did you use, what method did you apply, and what key insights did you gain or hope to gain from it?
Between 2018 and 2022, more than 60,000 customer comments were written, which served as the basis for this work. As these comments were not categorised in advance, the main objective was to automatically identify the topics and assess their relevance to customers. To do this, I used topic modelling, a form of unsupervised machine learning, to extract patterns and key themes from large collections of text data. The insights gained enabled a deeper understanding of customer opinions and needs. Of particular note was the ability to combine these results with available structural data such as age, gender, timing and purpose of travel to gain deeper insights into different customer groups.

How can your insights help our society?
Topic modelling offers the potential to process information more efficiently by automatically recognising patterns in large textual data. This allows people to focus on their strengths, such as creative thinking, collaboration and strategic decision making, rather than spending valuable time on manual data processing. At a societal level, this helps to free up resources for innovation-enhancing activities and social collaboration, ultimately contributing to a more advanced and efficient society.

How would you like to pursue your project in the future?
The Master’s thesis focused on a limited data set to demonstrate the possibilities and limitations of the method. The aim now is to transfer the developed approaches to operations. Specifically, new customer feedback will be analysed on a daily basis. The aim is to identify trends and new customer concerns more quickly and to develop appropriate measures, always with the aim of strengthening customer orientation.

How did your studies influence the project?
My studies have had a crucial influence on this project. Without the skills I acquired during my studies, I would not have been able to carry out this work. The modules of the MSc Applied Information and Data Science not only taught me basic skills such as programming in Python or working with databases, but above all showed me how to deal efficiently with different challenges. Looking back, the MSc was perfect for filling my knowledge gaps, especially in the technical area, and preparing me for a career as a data scientist.

What advice would you give to others starting similar projects?
Try to get access to the data you need as early as possible and focus on a clear research question. Allow plenty of time for data preparation, often much more than you think. I was particularly fortunate to be accompanied by an expert who provided very valuable technical and methodological input and took the time to scrutinise my interim results. With the support of a professional contact person in the company who understands the challenges, the Master’s thesis can become something that doesn’t disappear into a drawer, but brings real added value to you and the company.

And now for the end: What new hashtag are you aiming for in the future (e.g. for next year)?
It will stay the same next year, but with a bit more free time for other hobbies and projects :).

#StayCurious
#LifeLongLearning
#WhySoSerious?

We would like to thank Dominik Finzer for his dedication and time in sharing this wonderful research project with us. 

Decision Scientist - Internal Audit Manager/Data Analytics Leader - Data Engineer/Data Scientist - Analyst/Credit Management Office - Data Analyst - Senior Consultant - Specialist Data Analytics - Controlling Professional and Data Science Specialist - Data Science Consultant - Chief Digital Officer - Research Associate - Digital Solutions Engineer - Chief Revenue Officer - Malatec