Fair Access to Food Data in Africa: An Approach Based on Retrieval-Augmented Generation

In this paper, we propose a Retrieval-Augmented Generation-based (RAG) approach for fair access to food data in developing countries. This work also contributes to achieving sustainable development goals, specifically goal 2: Zero hunger and goal 3: Ensure healthy lives and promote well-being for all at all ages. Actually, given that a lot of African food data is accessible only in PDF format, we firstly extracted and organized these data using the Open Research Knowledge Graph (ORKG) and the image data using the Firebase database, with the link to the corresponding food description in the ORKG. Currently, more than 1000 foods eaten in around 50 African countries are already documented. Given that Large Language Models necessitate a lot of data and resources to be train/fine-tuned, which we do not have in our university, our idea is to use the food data stored in the ORKG to build a Retrieval-Augmented Generation system (RAG) for improving access to food data in Africa. Its implementation involved the use of multi-embedding models, vector databases (Pinecone), Falcon3 and LangChain. Because we do not have enough resources, this work is made possible thanks to a collaboration with TIB-Hannover who provide access to remote computers.