Applying Natural Language Processing to FEWS NET Reporting

Project Background

The International Center for Tropical Agriculture (CIAT), part of the Consultative Group on International Agricultural Research (CGIAR), are conducting a research project with USAID FEWS NET titled Food insecurity trends in the Famine Early Warning Systems Network (FEWS NET): An integrated machine learning approach to systematically analyze FEWS NET’s archive for historical trends and insights.

Approach

The periodic country analyses elaborated by FEWS NET have enabled the creation of a comprehensive historical database comprising consistent triannual reporting on wealth segmented Groups for sub-national Livelihood Zones in over two dozen countries. The longitudinal breadth of this Dataset has the potential to provide information and insights into long-term trends regarding shocks over time, coping capacity, and livelihoods. Systematically assessing the documentation from a historical perspective not only provides a comprehensive overview of food insecurity in FEWS NET covered countries, but can also support current operations by providing another evidence-base from which to carry out analyses and develop projections.

In such cases, machine-driven content analysis techniques stand as effective approaches to make sense of all the data available, as extracting knowledge from extensive text-based sources through conventional approaches such as manual coding or keyword searches is a challenge. In particular, text mining is an approach that seeks to automate the retrieval of high-quality information from text, usually by finding patterns and trends through machine learning, statistics, and linguistics. The aim of this technique is to enable the analysis of large amounts of unstructured text to derive insights. 

Research Questions

1. What historical trends (or anomalies) can be identified regarding the prevalence of various dimensions from the FEWS NET framework for food insecurity? I.e. reported hazards/shocks and outcome/impacts, particularly focused on Indicators related to infectious diseases, agroclimatology, markets, governance, and conflict

2. How do trends compare by region and type of Indicators?

3. How are the dynamics, such as the interactions within and between the different dimensions, represented? How do these representations transform in time and space?

Outcome

The outcome of this project will be an interactive Dashboard with analysis and Visualizations available to the FEWS NET team, food security community, and general public.