Constantino et al., "Representing the Disciplinary Structure of Physics" (2025)

Isabel Constantino, Sadamori Kojaku, Santo Fortunato, and Yong-Yeol Ahn, Quantitative Science Studies 6, 263–280 (2025)
DOI | arXiv

@article{constantino2025representing,
    author = {Isabel Constantino and Sadamori Kojaku and Santo Fortunato and Yong-Yeol Ahn},
    title = {Representing the Disciplinary Structure of Physics: A Comparative Evaluation of Graph and Text Embedding Methods},
    journal = {Quantitative Science Studies},
    volume = {6},
    pages = {263--280},
    doi = {10.1162/qss_a_00349},
    year = {2025},
}

Do graph and text embeddings capture consistent or divergent aspects of scientific structure? We test both methods using the Physics and Astronomy Classification Scheme (PACS) from American Physical Society publications. Neural network-based methods outperform traditional methods, and graph embedding methods such as node2vec are better than others at capturing the PACS structure.