An open-source platform to collect, process and visualise rescue data during Morocco's 2023 earthquake. Find our webapp hosted in HuggingFace and our poster in NeurIPS.
This repository contains some functions used in the processing of arabic and french user inputs, their analysis and visualization.
Examples of tasks:
- classifying user needs and NGOs supplies to predetermined categories using multilingual-e5 model
- mapping multilingual town names from raw users inputs to a standardized reference list using phonetical representation and text similarity methods
- an attempt to resolve the problem above using the LLM Mistral 7B
On September 8, 2023, a devastating magnitude 6.8 earthquake hit Morocco’s High Atlas Mountains, uniting Morocco’s government, Non-Governmental Organizations (NGOs), and citizens in an inspir- ing display of solidarity. Recognizing the need for improved relief coordination, our data-driven platform was created as a centralized hub to consolidate vital earthquake data and relief efforts.
Consolidating diverse, heterogeneous data sources is complex, particularly in low-resource languages like the Moroccan dialect. We employed Natural Language Processing (NLP) techniques to convert collected data into a refined and usable dataset. Key challenges included authentic data collection in crises, accurately identifying similar-named rural villages (‘douars’), and obtaining precise geolocation despite inaccuracies in mapping APIs (e.g. Google Maps and OpenStreetMap). Overcoming these obstacles was essential for effective humanitarian support.
More than 50 moroccan researchers, engineers and PhD students contributed to this work. You can find some of them in the contributors list in Nt3awnou's Hugging Face space.