SciDaSynth: Interactive Structured Data Extraction From Scientific Literature With Large Language Model
Problem
Extracting structured, standardized data from the explosion of scientific literature across text, tables and figures is slow and error-prone.
Why it matters
Evidence-based decision-making depends on efficiently aggregated, accurate data from many documents.
What the AI did
SciDaSynth, an interactive system powered by large language models, generates structured data tables from user queries by integrating text, tables and figures, with validation and semantic grouping to resolve inconsistencies.
Results
Within-subjects study with nutrition and NLP researchers showed higher-quality structured data more efficiently than a baseline; numbers not stated in source.
Impact
Evidence supports a working prototype validated in a user study, without quantitative benchmarks.
Record
| Venue | Campbell Systematic Reviews |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Xingbo Wang, Samantha Lee Huey, Rui Sheng, Saurabh Mehta, Fei Wang |
| Citations | 12 (OpenAlex, at import time) |
| References | 33 |
| Model or method | large language models (specific model not stated in source) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1002/cl2.70073 |
| Open access | diamond |
| Keywords | large language models, structured data extraction, scientific literature, systematic review |
Importance rating
significant — thousands to millions affected
Abstract states literature explosion makes structured data extraction critical for advancing scientific knowledge and evidence-based decision-making; broad cross-science relevance.