Data extraction evaluation toolkit

The Data Extraction Evaluation Toolkit (DEET) is an open-source toolkit designed to support the development and evaluation of AI-assisted data extraction for evidence synthesis.

Extracting structured information from research studies is one of the most time-consuming stages of the evidence synthesis process. While large language models (LLMs) and other AI methods have the potential to reduce this workload, inaccurate extraction can introduce errors into reviews and ultimately affect evidence used for research and policy decisions. Robust evaluation is therefore essential to ensure that AI tools are reliable, transparent, and fit for purpose.

DEET provides an end-to-end, modular workflow for configuring, running and evaluating automated data extraction methods. It enables users to compare AI-generated outputs against human-created reference (gold standard) datasets, helping to assess the accuracy, consistency and performance of different approaches. Designed around open and FAIR (Findable, Accessible, Interoperable and Reusable) principles, DEET supports reproducible evaluation through reusable code, APIs and standardised data models.

DEET is intended for researchers, evidence synthesis teams, developers and organisations interested in evaluating or developing AI-enhanced evidence synthesis workflows. Users can expect structured extraction outputs, performance metrics and transparent evaluation results that support informed decisions about the use of AI in evidence synthesis.

Example of how DEET is used

Within DESTINY, DEET will be used to evaluate an AI-powered data extraction workflow for intervention studies. Researchers will compare data extracted automatically by a large language model with manually extracted reference data, identify where the AI performs well or makes errors, and use these results to improve extraction methods before they are incorporated into evidence synthesis workflows.

Tool and documentation

Tool: https://github.com/destiny-evidence/data-extraction-evaluation-toolkit.git

Documentation: https://destiny-evidence.github.io/data-extraction-evaluation-toolkit/

People

Max Callaghan

Co-investigator, lead for evaluation

Potsdam Institute for Climate Impact Research

Ailbhe Finnerty Mutlu

Researcher

University College London

Diana Danilenko

Researcher

Potsdam Institute for Climate Impact Research

Ishita Gopal

Researcher

Potsdam Institute for Climate Impact Research

Kaitlyn Hair

Researcher

University College London

Kinlabel Okwen Tetamiyaka Tezok (Miyaka)

Researcher

eBASE (Effective Basic Services) Africa

Nadia Soliman

Researcher

University College London

Nik Loynes

Researcher

University College London

Sagar Uprety

Researcher

University College London

Tim Repke

Researcher

Potsdam Institute for Climate Impact Research

Yuhong Wang

Researcher

University College London