Spatial Information and Artificial Intelligence for Improving Food Security Monitoring Using Text Data

Context and challenges

This study focuses on food security, a major societal issue, and examines the early detection of food crisis risks in order to improve the effectiveness of the warning systems used by various governments and NGOs. Monitoring these crises using text sources remains largely unexplored, unlike the monitoring of epidemics, which has been driven by recent global health crises. Nevertheless, this poses significant methodological challenges related to the representation of information, such as the need for a detailed semantic analysis of spatial—as well as temporal—information through the representation of their “scope.” More specifically, the concept of scope refers to the spatiotemporal impact of food crisis risk factors that are not isolated events and that may be imprecise. It also applies to isolated events such as a flood or a pest infestation, whose effects on agricultural production—and thus on food security—are delayed in time or even in space. The spatiotemporal representation of food crisis risk factors therefore requires innovative contributions and represents a frontier in methodological science.

Goals

The proposed thesis falls within the field of early detection and monitoring of food crises, drawing on text information from informal sources such as news articles. The primary objective is to extract and appropriately represent the spatiotemporal characteristics associated with this textual data to improve the identification of related events. Taking into account the complexity of spatial information, including elements such as hierarchical and proximity relationships, is a current challenge for which there are few satisfactory solutions. The project focuses on developing a methodology that integrates spatial knowledge graphs into models derived from artificial intelligence (including generative models), with explicit consideration of spatial and temporal information. The proposed methods will be evaluated on an existing body of documents covering news from several West African countries over the past several years. A quantitative and qualitative analysis of the representation of the food crises under study will be conducted in order to extract causal factors and identify specific features associated with language or geographic region.

INRAE structure

Non-INRAE partner