fig1

Figure 1. Bibliometric analysis of materials information extraction (2002-2026): the paradigm shift toward generative AI. (A) Annual publication counts by assigned publication year (2002-2026). Papers published in 2021-2025 account for 73.6% of the total. Post-ChatGPT, annual publications surged threefold (from 70 in 2022 to 227 in 2025). The 2026 bar represents an incomplete publication year at the retrieval cutoff (26 January 2026) and is excluded from complete-year trend comparisons. The red curve is a cubic-spline smoothing of total annual publication counts and is shown only as a visual guide, not as a statistical model. Inset: pie chart showing the proportion of the entire retrieved corpus (assigned publication years 2002-2026) that utilizes LLMs; (B) Evolution of extraction methodologies. The share of LLM-based methods rose sharply from 3% in 2021 to nearly 50% in 2025, indicating a rapid transition to the generative AI paradigm. Search queries, data, and analysis scripts are available at https://github.com/sShuaiYang/AI4Mat_KnowExtraction. AI: Artificial intelligence; LLMs: large language models.