5 November 2020

Interview : AI & NLP for the Market Research industry

Data & Statistics See all news

How can Artificial Intelligence, Machine Learning and Natural Language Processing be an asset for collecting better data and getting better insights? At Gide, this topic matters a lot to us. That’s why we recruited Nour Medjedel, an Artificial Intelligence Engineer. In this interview, Nour talks about her career and her missions at Gide.

Machine Learning and NLP, two fields of expertise that make AI applicable

Hello Nour, first of all, what is your background?

I started my studies with a Bachelor’s degree in Computer Science at the University Lyon 1.  In my Master’s degree, I chose to specialize in Artificial Intelligence because from the beginning, it was a topic that fascinated me.

AI is very broad, sometimes obscure. What interested me the most is the link between AI and Data Science, because it is constantly evolving, and it is a field of concrete application of Artificial Intelligence. That’s why I chose to do my internship in the field of Machine Learning and NLP (Natural Language Processing). NLP covers all the technologies that allow machines to process and understand human language. 

Analyzing open-ended responses on large scale surveys

During your graduation internship at Gide, what subjects did you work on?

I worked on 2 different projects, with a common goal: to analyze open-ended responses on large scale studies, with a focus on sentiment analysis – i.e. how to identify the polarity or tone (positive or negative) of each response.

The first project was a survey with 280,000 responses. The responses were anonymous and open-ended, so the data collected were mainly text.

After cleaning and processing the data, I was able to automatically assign a positive or negative tone to each verbatim, using a pre-trained model on a very large corpus of data. I was also able to extract the most frequently used words.

The second project was a satisfaction survey on the level of service offered in store, in two languages (French and Dutch). The first step was to design an automatic language detection program. Then, I set up a supervised learning model : I used the satisfaction score to create a tone detection model. I took the time to test several models, to understand the most suitable method for this kind of project.

What technologies and tools do you use?

I work with the Python language, the most advanced in Data Science, especially in the field of Machine Learning. I use libraries, Open Source tools dedicated to data analysis and processing. I also rely on research published by specialists.

What Machine Learning and NLP can bring to Social and Market Research

Then you have been recruited on a permanent contract at Gide as an Artificial Intelligence Engineer. What can your work bring to Gide’s clients?

First of all, when I arrived at Gide, I received intensive training to discover the social and market research industry, and to understand how Gide teams work.

I discovered that open-ended responses are often overlooked in the analysis of survey data, even though they are a mine of information.

Until now, Gide’s clients have been collecting data from these open-ended questions, but they had three options for dealing with them:

  • Read all the answers and set up a manual classification (post-coding), which is impossible for large volumes of data ;
  • Use a service provider other than Gide for the analysis of this data;
  • Do not analyze this data at all.

The value of my work at Gide is to offer our clients a single point of contact, a single interface for handling and analyzing qualitative and quantitative questions in their surveys.

A lot of insights can be extracted from the open-ended responses (verbatims) and it would be a shame not to exploit them. We are therefore working on the creation of a visual interface allowing Gide’s clients to manipulate the textual data themselves.

In addition, my work will allow me to improve CAWIbot, our chatbot dedicated to the social and market research industry, so that it becomes even “smarter“. The challenge is interesting because this time, I’ll have to analyze a verbatim in “live” mode. The goal will be to make a verbatim match a list of expected answers, or to extract the topics discussed and their tone, and to lead the conversation.

A tailor-made approach adapted to the market research industry

What is the advantage of this way of working, compared to the existing text mining software?

At Gide, everything we do is custom-made, even though we rely on the state of the art and existing research. This means that we are able to deliver more accurate, customized results than NLP software can provide. And our experience in the market research industry allows us to provide results that are truly tailored to the needs of the market and social research professionals, and to the specific context of our client or project.

What do you like about Gide?

What interested me when I went to work at Gide was the ability to process textual data and work with large volumes of data.

But it was also the opportunity to interact with other team members and work with other tools: datavisualization, web tools, new technologies. That’s what I’m most passionate about: being able to be at the heart of innovative projects, in collaboration with other people, so that we can learn from each other.

Since my arrival at Gide I have been taken care of by an amazing team. The communication remains efficient and friendly despite the somewhat unusual conditions at the moment. It’s a pleasure to work with competent, dedicated and passionate people, each in their own field of expertise.