#44 Paris Women in Machine Learning & Data Science: Churn prediction, LLMs and tips to find a job as a foreigner

WiMLDS Paris
6 min readOct 3, 2023
Jihane, Caroline, Juliette, Joëlle, Laure, Natalie, Marie, Anna, Chloé

On September 27, our event at Ecole Normale Supérieure was a success: the room was fully booked.

Our volunteer Natalie moderated the event and shared a new step for our community: we just reached 5,000 members! To celebrate, we offered some WiMLDS teeshirts at the end of the event.

Natalie then presented a workshop on Ethical Artificial Intelligence, coorganised by Ruta Binkyte, one of our past speakers. It will take place on November 23th-24th 2023.

Joëlle Lautré

Our first speaker, Joëlle Lautré, Client knowledge and Data science manager at Bouygues Telecom, presented a talk on churn prediction within the telecommunications industry.

Bouygues Telecom has access to a wide array of data, collected with customer consent, including customer information, TV audiences, billing, customer service calls.

Churn prediction, one among many use cases her team addresses, was the central focus of Joëlle’s presentation.

There are two scenarios in which churn prediction is vital:

  • When customers contact customer service, it is essential for the operator to tailor their response based on churn risk. They employ a classical scoring method, using gradient boosting techniques (Light GBM), resulting in a calculated churn probability for each customer.
  • A growing number of customers are ceasing contact support before canceling their contracts. Thus, Bouygues Telecom must proactively engage with these customers before they cancel their contracts. In this case, the approach involves target detection using a sequence mining algorithm (SPADE). The objective here is to identify and engage with high-risk customers during their journey.

In both scenarios, one of the primary challenges faced was achieving a low false positive rate to ensure that generous loyalty rewards were not incorrectly offered.

However, achieving better predictions made customer retention more challenging, as the identified customers had already made their decision. Thus, a balance needed to be found between the volume of targeted customers and the concentration of churn-prone customers.

The model is now in production, automatically generating new targets within CRM tools. Congratulations to the team on their achievement!

If you want to know more, check out her slides below.

Laure Soulier

Our second speaker, Laure Soulier, an associate professor at Sorbonne University, is an expert on representation learning, natural language processing (NLP) and information retrieval. She has advised several theses in these domains. In her presentation, Laure talked about a very hot topic in the community at the moment with the ChatGPT phenomenon: how to supervise a PhD thesis in NLP in the age of Large Language Models (LLMs)?

Laure started with an overview of transformer networks, emphasising their ability to model words within their context. An interesting aspect of LLMs lies in their capability to facilitate in-context learning through the formulation of well-crafted prompts (instructions to the generative model). This allows for learning from examples mentioned in the prompts without the necessity of fine-tuning. Of course, for specialised tasks, fine-tuning a generic model remains necessary.

Given the immense power of these models, Laure explored their impact on NLP research. She shared insights from a project she had supervised a few years ago, focused on conversational search systems. Back then, traditional search engines like Google Search dominated the landscape. The project’s objective was to transform a conventional search engine into a conversational search system. Laure’s student’s thesis addressed the comprehension of user information needs when expressed in natural language. Achieving this required the implementation of a multi-turn interaction framework to enable the system to iteratively refine its understanding of user queries.

But what do you do when ChatGPT-3 and similar models emerge, outperforming the work you’ve invested in during your thesis? Laure offered a pragmatic approach: sidestep the competition and explore ways to incorporate LLM models into your project, providing supplementary tools or perspectives by integrating them.

She underscored the importance of agility for Ph.D. students, recognising that competing against major players with substantial GPU clusters is a complicated challenge. Instead, staying abreast of the latest literature and adapting swiftly are key strategies. A thesis should not be seen as a rigid, three-year project anymore.

Laure concluded her presentation on an encouraging note for Ph.D. students: “Don’t be afraid, you don’t need to create a super transformer. You’ll find good ideas, contribute to the community and will have learned a lot.”

If you want to know more, check out her slides below.

Anna Abreu

The third presentation, which typically leans towards a less technical aspect, featured Anna Abreu, a Data Analyst at Amazon, who was the perfect speaker to guide our international community on the topic of finding a job in France as a foreigner.

Having previously worked as a data analyst for a year, Anna realised that in order to continue pursuing this career in France, she needed to get a Master’s degree. This marked the beginning of her journey into French culture.

She shared her personal experiences as a foreigner seeking employment in France, highlighting the various realisations she had, including the challenges and cultural differences she encountered.

She touched upon several key points, emphasising the importance of academic qualifications, learning French, and offering insights on where and how to acquire French language skills. Anna also shared valuable information on where to find companies that are open to hiring English-speaking employees (she created a useful list available here), discussed salary expectations, administrative prerequisites, and provided guidance on crafting a CV that aligns with French standards.

Anna strongly stressed the importance of being prepared for administrative requirements.

In addition to practical tips, she recommended a valuable resource for better understanding French culture: Erin Meyer’s book, “The Culture Map.” This book is a valuable tool for gaining insights into the nuances of working and living in France.

If you want to know more, check out her slides below.

If you do not want to miss our our events, you can:

🔗 follow us on Twitter, Meetup, and LinkedIn

📑 check our Google spreadsheet if you want to speak 📣, host 💙, help 🌠

📍join our Slack channel for information, discussions, and opportunities

📩 send an email to the Paris WiMLDS team to paris@wimlds.org

🎬 subscribe to our WiMLDS Paris Youtube channel

📸 follow the global WiMLDS on Instagram, LinkedIn, and Facebook

🔥 share your company or lab’s job offers for free on the global WiMLDS’ website.



WiMLDS Paris

WiMLDS Paris is a community of women interested in Machine Learning & Data Science