Numbers are the language of science—but in research articles, they’re frequently buried within the text content and difficult to analyze. Researchers at Jülich have evolved an AI system that automatically detect these numbers, categorizes them, and transforms them into structured data. The Quinex framework for that reason removes the need for time-consuming manual work.
Whether in energy, climate, or material research—scientific papers are full of numbers—or, extra exactly, quantitative data: efficiencies, temperatures, costs, emissions. These are often vital for enhancing models or detecting trends. At the same time, the number of scientific publications is growing hastily. For many research questions, it is now virtually impossible to manually evaluate all appropriate publications—the time and resources needs would be massive.
The Quinex (“Quantitative Information Extraction”) framework, evolved via researchers at Jülich, is based totally on language models and simplifies this procedure: Artificial intelligence detect numerical values, assigns them to suitable units, and identifies what was measured, while, where, and how. Thus, a sentence like “Efficiency level of 63 to 71% are considered for 2025” is converted into a structure dataset including all appropriate contextual information—from the year and measurement method to the source.
Open and Efficient AI
Unlike many proprietary AI solutions, Quinex is primarily based completely on open, enormously small, and therefore efficient language models. These have been particularly trained to detect and sort quantitative information in scientific texts. Compared to similar systems, Quinex can provide more specific results, collects contextual information in a more nuanced manner, and also takes implied traits under account.
In spite of its compact size, Quinex obtains a recognition accuracy (F1) of about 98% for numbers and related units, and about 87% and 82% for the classification of quantified properties and entities. These high accuracy rates were done by particular generated training datasets and methodological improvements.
“We wanted to design a tool that is powerful, yet also transparent and resource-efficient,” explains Dr. Jann Weinand, head of the incorporated Scenarios Department at Jülich System Analysis. “Quinex makes artificial intelligence more accessible for data analysis in science.”
Successful practical test
To test Quinex’s practical suitability, the system was applied to thousands of scientific simplifies from numerous fields. It effectively extracted data on electricity manufacturing costs for numerous energy technologies, on maximum oxygen uptake in humans, on earthquake magnitudes and locations, and on the band gaps of photovoltaic materials.
The automatically derived values carefully matched the respective reference data. This shows that Quinex is properly-suitable for analyzing huge volumes of academic literature across a wide range of research fields and deriving dependable traits from it.
New perspectives for research
“Language models open up latest views for science and assist preserve an overview of complete research fields,” stated lead author Jan Göpfert. “They allow automated literature searches, the creation of uniformly structured research databases, and trend analyses that shows development in science and technology at an early stage.”
“Our goal is to alleviate researchers of routine work,” stated Dr. Patrick Kuckertz, head of the Research Data Management Group. “Quinex is designed to support them arrive at insights more quickly and manage the growing flood of data in science.”
Limitations and future upgrades
Quinex isn’t always entirely error-free either—however transparency is part of its design. “The system acknowledges numbers and units very reliably,” stated Göpfert. “Since they’re taken directly from the text, they can’t be ‘hallucinated.’ Moreover, misinterpretations sometimes occurs, as an example, whilst crucial references are distributed throughout the text.”
Thus, Quinex stays a tool that helps people but does no longer replace them. “We propose using Quinex in which it informs and relieves researchers—but the responsibility for clarifying the results stays with them,” stated Göpfert. Every recognized number may be traced returned to its source and, wherein feasible, is emphasized in the original text.
The team is working to similarly expand Quinex with additional domain-specific datasets and models, making it even more efficient and flexible enough to adapt to numerous research requirements.
Open collaboration welcome
Forschungszentrum Jülich is making Quinex avaliable as an open-source venture. This is planned to present researchers global the opportunity to test, amplify, and adapt the system to their own fields—from energy research to chemistry and biomedicine.
