Reading Between the Posts : Evidence from Facebook Expat Groups

Imagine you’ve just arrived in a new country, for studies, for work, or simply taking a shot at life. You don’t speak the language fluently. The landlord you had a ‘solid’ agreement with suddenly backs out. To rent another place, you need a bank account, but to open it, you need proof of housing. You call the social security office, but they speak too fast. You’re lost in administrative translation.

So who do you turn to? For many immigrants, the answer is: strangers in Facebook groups.

From visa questions and housing leads to everyday discussions, immigrant support communities online, or ‘expat groups’, have become an essential part of how people navigate life abroad. These groups are practical, decentralized spaces, often organized by nationality (“Colombians in Lille,” “Americans in Lyon,” “British in Paris”), covering a wide spectrum of topics, tones, and intentions. For some, it’s about making ends meet; for others, it’s about building a life that feels truly their own.

Unlike most studies based on surveys or administrative records, this project draws on bottom-up, unfiltered narratives from immigrants themselves. Our aim is to use these digital traces to understand what drives people to seek help in online support groups, and how they articulate their needs.

We analyze over 23,000 posts, covering around 20 different nationalities, collected from over 84 expat groups. With the help of large language models (LLMs), in addition to traditional natural language processing techniques, we identified surface-level patterns (such as country-specific vocabulary), extracted self-declared traits (such as the poster’s gender or housing budget), and detected deeper narrative signals including urgency, family references, or expressions of religious identity. We then relate these features to country-level traits (such as GDP per capita or EU membership) to explore broader differences in how the various expat communities describe and approach life in France.

Read more ...