Developing an AI-powered tool for data extraction from texts

SEI champions systematic evidence syntheses (including systematic maps and reviews) to support evidence-informed decision-making. A comprehensive synthesis of available research is invaluable to understand complex societal challenges, for example related to the environment and development. It can however be quite costly and time consuming if the synthesis involves a large body of research.

In this project, we are developing and testing an AI-powered and expert-validated process for data extraction from texts. The aim is to enhance the efficiency and comprehensiveness of systematic evidence syntheses, policy analyses and similar methodologies that apply text analysis. If proven reliable and sufficiently precise, this AI tool could improve decision-support capabilities by efficiently providing more accurate evidence.  

The proposed approach applies Large Language Models (LLMs) with semantic search to extract information buried in documents. This makes it possible to identify answers to questions through meaning (rather than relying on keyword search). The AI-driven semantic coding is then combined with human expert validation.

In this collaboration between SEI and the KTH Royal Institute of Technology in Stockholm, the AI-powered data extraction process is piloted on a review of how well Water, Sanitation and Hygiene (WASH) policies integrate climate policy objectives, and vice versa. To ensure diverse perspectives and domain-specific knowledge, a pool of subject experts from various SEI centres will validate the AI outputs.