Email this article Automated reconstruction of literary character social roles using graph neural networks: a multilingual corpus approach
- Authors: Dragomirov D.S1
-
Affiliations:
- Saint Petersburg State University
- Issue: No 8 (2025)
- Pages: 129-136
- Section: Articles
- URL: https://journals.rcsi.science/2541-8459/article/view/371256
- ID: 371256
Cite item
Abstract
an approach for quantitative reconstruction of literary characters' social roles using graph neural networks is developed. A multilingual corpus (~40 works) of classical prose in Russian, Chinese, Japanese, and Korean is compiled. Automated processing includes character graph extraction and application of Graph Neural Networks (GCN, GAT) for role classification. Integration with a large language model in GraphRAG scheme provides result interpretation. The graph model correctly identifies central characters with 92% accuracy. Cross-cultural differences are revealed: Russian novels demonstrate centralized networks around the protagonist, while East Asian texts show more distributed structures. Speech form analysis revealed correlation between character status and honorific usage (?, ?, "Вы"). The approach is applicable for digital hermeneutics, educational applications, and publishing analytics. For the first time, cross-cultural quantitative analysis of literary networks using GNN is performed.
References
- Pang N., Sun M., Zhu H. Louise or Ferdinand? Exploring the protagonists of Love and Intrigue using social network analysis // Digital Scholarship in the Humanities. 2023. Vol. 38. № 3. P. 1214 – 1232. doi: 10.1093/llc/fqad007
- Eberle O., Gleave J., Schroeder C. et al. Historical insights at scale: a corpus-wide machine-learning analysis of early-modern astronomic tables // Science Advances. 2024. Vol. 10. № 43. Art. eadj1719. P. 1 – 16. doi: 10.1126/sciadv.adj1719
- Dadlani A., Yu M., Roberts S. et al. Leading by the nodes: a survey of film-industry network analysis and datasets // Applied Network Science. 2024. Vol. 9. Art. 76. P. 1 – 23. doi: 10.1007/s41109-024-00673-9
- Yang X., Zhang Z. Enhancing book genre classification with BERT and InceptionV3: a deep-learning approach for libraries // PeerJ Computer Science. 2025. Vol. 11. Art. e2934. P. 1-20. doi: 10.7717/peerj-cs.2934
- Schmidt D., Puppe F. Automatic extraction and visualization of interaction networks for German fairy tales // Machine Learning and Knowledge Extraction. 2024. Vol. 6. № 4. P. 2447 – 2493. doi: 10.3390/make6040121
- Sugishita K., Masuda N. Social network analysis of manga: similarities to real-world social networks and trends over decades // Applied Network Science. 2023. Vol. 8. Art. 79. P. 1 – 17. doi: 10.1007/s41109-023-00604-0
- Zhao W., Liu Y., Han J. A block-based adaptive decoupling framework for graph neural networks // Entropy. 2024. Vol. 24. № 9. Art. 1190. P. 1 – 22. doi: 10.3390/e24091190
- Zhang C., Wang S., Zhan D. et al. Inferring users’ social roles with a multi-level graph neural network model // Entropy. 2021. Vol. 23. № 11. Art. 1453, P. 1 – 20. doi: 10.3390/e23111453
- Mitheran S., Java A., Sahu S. K., Shaikh A. Introducing self-attention to target-attentive graph neural networks // arXiv preprint arXiv:2107.01516. 2021. P. 1 – 4.
- Fatemi B., Halcrow J., Perozzi B. et al. From local to global: a Graph-RAG approach to query-focused summarization // arXiv preprint arXiv:2404.16130. 2025. P. 1 – 26.
- Jou E. Honorification as Agree in Korean and beyond // Glossa: Journal of General Linguistics. 2024. Vol. 9. № 1. P. 1 – 29. doi: 10.16995/glossa.9565.
- Knollmeyer S., Caymazer O., Grossmann D. Document Graph-RAG: knowledge-graph-enhanced retrieval-augmented generation for document QA within the manufacturing domain // Electronics. 2025. Vol. 14. № 11. Art. 2102, P. 1 – 19. doi: 10.3390/electronics14112102
- Gupta S., Sharma R., Masuda N. Influence maximization on temporal networks: a review // Applied Network Science. 2024. Vol. 9, Art. 16 (= 65). P. 1 – 25. doi: 10.1007/s41109-024-00625-3
Supplementary files
