Evaluation framework related to metadata extraction from scientific publications

This dataset provides the evaluation framework used in a systematic literature review of methods for extracting and aligning data from tables and charts in scientific publications. The evaluation framework captures structured annotations for 68 peer-reviewed studies, covering tasks, modalities (tables, charts, multimodal), benchmarks, models, architectures, evaluation metrics, and support for variable identification, alignment, and reconstruction.

The evaluation framework is designed to support reproducibility, comparative analysis, and meta-research on multimodal document understanding. It enables quantitative and qualitative analysis of trends, open challenges, and methodological gaps in table and chart extraction, including multimodal alignment, variable and value association, and benchmark reuse.

This resource can be reused by researchers developing extraction systems, benchmarking multimodal models, or studying the state of the art in scientific document analysis, particularly in the context of structured data extraction from tables and charts.