What question did this study set out to answer?

This research aims to improve the extraction and alignment of multimodal cultural heritage data.

Construction of multimodal knowledge graph for cultural heritage: enhanced joint extraction and domain-adaptive cross-modal alignment method

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial demonstrates enhanced extraction and alignment in cultural heritage data, suggesting improved digital humanities tools.

This research aims to improve the extraction and alignment of multimodal cultural heritage data.
Constructed a domain-standardized multimodal dataset for training and evaluation.
Proposed an enhanced joint extraction model BS-PTWF for extracting entities and relations.
Performed domain-adaptive fine-tuning on the Chinese-adapted CLIP model for cross-modal alignment.
Achieved an F1-Score of 82.5% on the self-constructed dataset, 4.8 percentage points higher than the baseline.
Attained Recall@10 of 97.4% in image-to-text retrieval tasks, indicating strong semantic alignment.
F1-Scores of 61.0% and 80.6% were reached on the CMeIE and DuIE2.0 datasets, respectively.

Song et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark