^{In their third year, bachelor's students begin the academic year with a minor, which offers them the opportunity to personalise their curriculum and explore areas beyond their regular programme. Whether through internships, courses from other faculties, earning credits toward a master’s programme, or diving into subjects they never expected to interest them, the minor allows students to expand their horizons. In this blog post, Silvia Sanguinazzi, a third-year Data Science & Society student, talks about her six-month internship at Wonderflow. Keep reading to find out how it all went!}

Hi there!

Hello! My name is Silvia, and I come from Milan, Italy. I am a third year Data Science & Society student with a passion for Natural Language Processing (NLP). I had the exciting opportunity to intern at Wonderflow from July to December 2024. In this blog post, I’ll share my journeywhat I worked on, the challenges I faced, and the insights I gained along the way. I hope my experience will inspire you! 

The Start 

Wonderflow is a leader in AI-powered consumer insights, helping businesses analyze and transform raw customer feedback into actionable insights. I was fortunate to join the Data Science and Linguistics team, where I got the chance to apply my academic knowledge in a real-world setting. My internship project had two main areas of focus: improving keyword extraction capabilities and exploring metadata extraction from unstructured text. 

I found the internship through contacts in my network. The timing of the internship was perfect, as I had just completed some foundational courses in NLP and was eager to dive deeper into the field. The company, with its headquarters location in Amsterdam, offered me the flexibility of working fully remote, which was a great fit for my situation.

Project 1: Keyword Extraction Optimization 

• The Challenge: When I joined Wonderflow, one of the first projects I worked on was enhancing the company’s keyword extraction pipeline. The goal was to improve the efficiency and accuracy of keyword extraction from customer reviews using advanced NLP techniques, and large language models (LLMs). 

• My Contribution: I initially assessed the existing keyword extraction system, which used a simpler NLP model. However, the results often produced irrelevant keywords or missed important ones. This led to a shift in the project’s direction, opting for a more sophisticated approach with LLMs, which provided higher-quality and cost-effective results. 

  I worked on fine-tuning prompts, cleaning text data (using techniques like lemmatization and regex), and integrating the improved extraction process into a new pipeline. 

• Funny anecdote: While analyzing animal food reviews, I came across some surprisingly honest feedback that made it trickier to draw conclusions. It wasn’t uncommon to find reviews where people joked about how they couldn’t imagine eating the food themselves, even though their dogs seemed to enjoy it. Some reviews were just so blunt about the experience, that it was difficult to determine whether the product’s quality was actually good or just tolerated by the pets. This added an extra layer of complexity to the analysis, as the feedback wasn’t always clear-cut, especially when it came to distinguishing genuine preferences from humorous exaggerations. 

• The Outcome: By the end of our work on this project, the new pipeline was substantially more efficient and cost-effective. The results not only improved keyword relevance but also saved the company money, which was a great win for both the technical and financial side!

❝The results not only improved keyword relevance but also saved the company money, which was a great win for both the technical and financial side!❞

Project 2: Extracting Metadata from Unstructured Text 

The Challenge: After the MPK project, I shifted focus to a new challenge—extracting metadata related to product usage, such as location, time, and activity, from customer reviews. This metadata was crucial for improving the company's ability to understand the contexts in which customers use products and derive deeper insights from their feedback. 

My Contribution: The first step involved defining the target metadata fields and determining the best methods for extracting them. I researched various extraction techniques and refined a prompt, using a large language model, which was chosen for its ability to handle unstructured text effectively. 

We designed different methods for metadata extraction, such as using separate prompts for each metadata type (location, time, activity) and combining extraction and classification into a single process. After several iterations and testing across different product categories, we found the most efficient method for better precision. 

Reflection: Through this project, I learned how essential it is to tailor NLP techniques to the specific needs of the data you’re working with. I gained invaluable experience in prompt engineering and testing, ensuring that the extracted metadata was both accurate and relevant to the product context. 

Key Takeaways from My Internship 

1. Practical Application of NLP Techniques 

   One of the most rewarding aspects of my internship was applying theoretical concepts to actual problems and seeing tangible results, which was incredibly fulfilling.

2. Collaboration is Key

   Working alongside experienced data scientists, linguists, and engineers taught me the value of collaboration. Every team member brought a different perspective, and together, we overcame challenges and implemented solutions that had a direct impact on the final product. 
   

3. The Importance of Continuous Learning

   The rapid pace of innovation in NLP meant that I had to keep up with the latest research and tools. This internship reinforced my belief that in fields like NLP, you must always be learning to stay ahead. 

❝Working alongside experienced data scientists, linguists, and engineers taught me the value of collaboration.❞

Why This Internship Matters 

In a nutshell, this internship allowed me to make meaningful contributions to an innovative AI company and gain valuable experience in some of the most exciting areas of NLP. 

This internship solidified my passion for NLP and gave me the tools and confidence to pursue a career in this rapidly-evolving field. I look forward to applying everything I’ve learned as I continue my studies and on future professional opportunities. 

❝This internship solidified my passion for NLP and gave me the tools and confidence to pursue a career in this rapidly-evolving field.❞

For prospective students, if you’re looking for an internship that offers both technical challenges and opportunities for growth, I highly recommend pursuing a role in a team like the one at Wonderflow. It’s a unique chance to apply what you’ve learned in your courses, learn new skills, and make an impact in the world of data science. 

Thanks for reading, and I hope you found this post inspiring! 

Join the programme and be next in line for an inspiring opportunity!

Visit our BSc Data Science & Society webpage and sign up for our monthly newsletter. See you there!