Multidimensional Information Extraction from Digitized Chinese Historical Documents via Relation-Aware Semantic Enhancement
DOI:
https://doi.org/10.54097/5pkh5531Keywords:
Chinese historical documents, optical character recognition, relation modeling, reading-order recovery, named entity recognitionAbstract
Complex layouts, rare characters, and irregular reading orders pose challenges to optical character recognition (OCR) and semantic information extraction from Chinese historical documents. To address these challenges, we propose Relation-Guided and Semantics-Enhanced TongGuOCR (RSE-TongGuOCR). A category-aware relation predictor integrates learned reading dependencies into geometry-based block construction, while dynamic programming selects recognition blocks to preserve both local context and character legibility. To promote consistency between layout analysis and text decoding, a cross-branch constraint aligns layout-based relation predictions with decoder-generated transition tokens in a shared spatial transition space. The framework retains rare-character vocabulary expansion and incorporates an ancient Chinese named entity recognition (NER) branch to identify six categories of historical entities in transcribed text. The OCR and NER components are trained in stages using distinct annotation sources and evaluated independently. RSE-TongGuOCR achieves an AR of 94.38 and an RO-ED of 2.86 on M5HisDoc, while its NER branch attains an F1 score of 85.84% on EvaHan2025 Dataset A. These task-specific results support the framework’s effectiveness in document transcription, reading-order recovery, and structured historical information extraction.
Downloads
References
[1] Zhou, Z., Sun, Y., He, H., et al. (2026). TongGuOCR: A layout-aware and token-augmented OCR MLLM for Chinese historical documents. arXiv preprint arXiv:2608.07917v2.
[2] Li, Z., Liu, Y., Liu, Q., et al. (2026). MonkeyOCR: Document parsing with a structure-recognition-relation triplet paradigm. arXiv preprint arXiv:2506.05218v2.
[3] Niu, J., Liu, Z., Gu, Z., et al. (2025). MinerU2.5: A decoupled vision-language model for efficient high-resolution document parsing. arXiv preprint arXiv:2509.22186.
[4] Cui, C., Sun, T., Liang, S., et al. (2025). PaddleOCR-VL: Boosting multilingual document parsing via a 0.9B ultra-compact vision-language model. arXiv preprint arXiv:2510.14528.
[5] Shi, Y., Peng, D., Zhang, Y., Cao, J., & Jin, L. (2025). A large-scale dataset for Chinese historical document recognition and analysis. Scientific Data, 12, Article 169. https://doi.org/10.1038/s41597-025-04897-8
[6] Zhang, Q., Duan, Z., Ma, S., Liu, S., Yuan, Z., & Ma, R. (2025). Multi-domain ancient Chinese named entity recognition based on attention-enhanced pre-trained language model. In Proceedings of the Second Workshop on Ancient Language Processing (pp. 237–241).
[7] Lu, Y., & Lei, M. (2025). Construction of NER model in ancient Chinese: Solution of EvaHan 2025 challenge. In Proceedings of the Second Workshop on Ancient Language Processing (pp. 165–169).
[8] Hua, W., & Xu, S. (2025). When less is more: Logits-constrained framework with RoBERTa for ancient Chinese NER. In Proceedings of the Second Workshop on Ancient Language Processing (pp. 192–196).
[9] Li, B., Chang, B., Liu, R., et al. (2025). Overview of EvaHan2025: The first international evaluation on ancient Chinese named entity recognition. In Proceedings of the Second Workshop on Ancient Language Processing (pp. 156–164).
[10] Hunyuan Vision Team, Lyu, P., Wan, X., et al. (2025). HunyuanOCR Technical Report. arXiv preprint arXiv:2511.19575.
[11] Duan, S., Xue, Y., Wang, W., et al. (2026). GLM-OCR Technical Report. arXiv preprint arXiv:2603.10910.
[12] Zhang, Z., Liu, H., Liang, S., et al. (2026). PaddleOCR-VL-1.6: Expanding the frontier of document parsing with under-optimized region refinement and progressive post-training. arXiv preprint arXiv:2606.03264.
[13] Wang, B., He, T., Ouyang, L., et al. (2026). MinerU2.5-Pro: Pushing the Limits of Data-Centric Document Parsing at Scale. arXiv preprint arXiv:2604.04771.
[14] Zhang, Y., Liu, M., Tang, H., Lu, S., & Xue, L. (2025). Simple Named Entity Recognition (NER) System with RoBERTa for Ancient Chinese. In Proceedings of the Second Workshop on Ancient Language Processing (pp. 206–212). https://doi.org/10.18653/v1/2025.alp-1.27
[15] Dong, W., & Liu, M. (2025). Multi-strategy named entity recognition system for ancient Chinese. In Proceedings of the Second Workshop on Ancient Language Processing (pp. 213–220).
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Academic Journal of Applied Sciences
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
