Automatic identification of refugee needs
We created a tool for the automated identification of refugee needs regarding the Ukraine Refugee Crisis. Our tool uses publicly available data from Telegram, a popular messenger service among the Ukrainian population. We observed on the platform that in most European countries, open groups formed e.g. “Refugees in Germany” where refugees communicate needs and solutions to common problems. We utilise state-of-the-art machine learning to identify clusters within the messages posted on the platform Telegram. We identified the top eight clusters, which are directly linked to problems refugees face in Europe. These eight problems are: medical care, accommodation, transport to and from Ukraine, government services and asylum, animal welfare and regulations, banking, public transport within Europe, and COVID-19 vaccinations. As we scraped data from 23 countries, we can observe how these problems arise within different countries and over the duration of the crisis spans. While our project is done by swiss researchers at the University of Zurich and supported by various public administrative organisations (see attached link to the project: “project website”), for the call of the DT4Regions we adapted our solutions to identify refugee needs in Europe. While we try to communicate some of the benefits our tool brings, we would appreciate, if you try it out yourself (see attached link to the tool: “automated identification of refugee needs tool”).
Problem or opportunity
As research has identified (e.g. https://www.tandfonline.com/doi/abs/10.1080/1369183X.2013.855074), refugee management utilises a top-down approach. It often focuses on border management and the economic cost of refugees rather than identifying the needs of the refugee population. Since the start of the Ukraine Crisis, Europe faces a challenging stream of refugees from Ukraine. However, it is unclear which problems refugees face in their host country. The administrative processes within the EU are different across all host countries. Our system allows monitoring the most challenging problems for refugees within Europe generally and further allows us to analyse each country on its own. We see that the problems of refugees vary over location and time. Generally, our tool helps with the problem of identifying refugee needs. For refugees, it tackles the problem of voicing the community's needs and supports refugees within Europe.
Contrary to the widely used top-down refugee management, our system allows for bottom-up management of the Ukrainian Refugee Crisis. Our tool is built on data from refugees for refugees and does not include the agenda of any political party. We thus provide the following functionality, beneficial for the refugee community:
- show refugee needs within Europe, but also in single countries.
- refugee needs are identified via multi-lingual clustering.
- possibility to look at individual clusters and compare them.
- define the time of interest.
- theoretically, the system can be deployed in near real-time.
- approach can easily be transferred to other crises.
- the first system that allows for bottom-up refugee management.
- from over one million messages, we can separate noise from important content for the refugee community.
Through the tool, we can make the following conclusions:
- Some countries have pioneer roles, needs seen within them can be observed later in other countries.
- Problems identified vary in intensity based on the country regarding absolute and relative values.
- Refugees are largely self-organised. Only a few telegram accounts are managed by public administration. However, as telegram is the main communication channel for refugees, they would profit from such offerings.
- Analysing telegram data, we can map common questions to solutions.
In general, we see that public administration, through our provided tool, can provide data-driven policies and act efficiently regarding the Ukrainian Refugee Crisis. The refugee community can profit thus from better public services across Europe.
We scraped data from 23 open Telegram groups used for the coordination of Ukrainian refugees utilising the Telegram API. Each group relates to one European country. With the obtained data set containing more than a million messages, we applied BERTopic modelling (https://maartengr.github.io/BERTopic/index.html). The algorithm is a state-of-the-art topic modelling technique that allows the clustering of unstructured text data. The model leverages the transformer architecture, allowing current breakthroughs within artificial intelligence. We are thus able to process and comprehend text from over 50 languages, including Ukrainian and Russian, spoken by the Ukrainian population. We set the algorithm to provide us with the 10 top clusters. Afterwards, we manually merged similar clusters and gave distinct clusters a name. After clustering the messages, we can identify and observe how the clusters vary over location and time.