EviCal: Evidence-Grounded Consistency Calibration for Content-Level Multimodal Labeling
Informatics · 11.06.2026 · 13 (6) · 86
DOI: 10.3390/informatics13060086 ↗Kurz gefasst
Fachartikel in der Zeitschrift Informatics, online erschienen am 11.06.2026. Verfasst von Xiaofeng Zhang und 6 weiteren Autor:innen. Die Arbeit nennt 49 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Science and Technology Project of State Grid Zhejiang Electric Power Co., Ltd..
Abstract (Originalsprache)
Power system testing and inspection documents are multimodal and highly structured, making content-level audit labeling challenging due to scattered evidence and cross-component dependencies. We propose EviCal, an evidence-grounded consistency calibration framework under a predefined label space. EviCal decomposes documents into atomic units (text segments, table rows, and figure captions), grounds each label to minimal supporting evidence via label-aware semantic focusing, calibrates local decisions against global causal and logical constraints imposed on symbolic intermediate states, and produces explicit confidence estimates. Experiments on two real-world power-system datasets show that EviCal achieves up to 93.97% accuracy and 81.22 F1, and attains a human score of up to 4.58/5, outperforming strong multimodal baselines and delivering more reliable, interpretable audit predictions.
Förderung
- Science and Technology Project of State Grid Zhejiang Electric Power Co., Ltd.
Zitieren
APA
Zhang, X., Han, B., Yuan, Y., Zhu, G., Song, H., Qiu, W., & Ni, L. (2026). EviCal: Evidence-Grounded Consistency Calibration for Content-Level Multimodal Labeling. Informatics, 13(6), 86. https://doi.org/10.3390/informatics13060086
Vancouver
Zhang X, Han B, Yuan Y, Zhu G, Song H, Qiu W, et al. EviCal: Evidence-Grounded Consistency Calibration for Content-Level Multimodal Labeling. Informatics. 2026;13(6):86. doi:10.3390/informatics13060086
BibTeX
@article{Zhang2026,
author = {Zhang, Xiaofeng and Han, Baoli and Yuan, Yufeng and Zhu, Guangyao and Song, Huibo and Qiu, Weixing and Ni, Li},
title = {EviCal: Evidence-Grounded Consistency Calibration for Content-Level Multimodal Labeling},
journal = {Informatics},
year = {2026},
volume = {13},
number = {6},
pages = {86},
doi = {10.3390/informatics13060086}
}
EUVIMED