A scientific paper is not just a collection of results: it's an imprint of research handwriting. Some scholars trace strict bibliometric lines, others add experimental flourishes, and still others inscribe content-analytic curlicues. Over 30 years, three top journals of library and information science accumulated 5281 such 'signatures' from 81 countries. In 1990, the average paper was a single-country autograph. By 2019, it was a collective signature spanning 1,45 countries.

To decipher this planetary handwriting, Chenzhi Zhang's group used machine learning, training a neural network called CogFT. It's based on SciBERT, a transformers model with an attention mechanism that understands the natural language of science. Like a graphologist who picks out key loops and strokes in handwriting, CogFT extracts four fragments from the full text that describe the method. Recognition accuracy for 15 method categories, from bibliometrics to surveys, reached 0,83 macro F1.

The methodological picture turned out to be varied. Belgium and the Netherlands use a bibliometric stroke in nearly half of their papers. China and Brazil prefer experimental pressure — more than a third of their papers. To measure diversity, the authors applied the Shannon index — the same measure ecologists use for species richness. H = -∑ $p_i$ log₂ $p_i$, where $p_i$ is the proportion of papers using method i. The higher H, the richer the arsenal. Over 30 years, almost all countries expanded their methodological alphabet, and only Belgium remained true to a limited set of letters. Cluster analysis using dimensionality reduction into a 2D map (UMAP) divided trajectories into three groups: 'mature' (USA, UK, Canada) with consistently high diversity, 'fast-growing' (China, Australia, and 14 other countries) with a sharp rise, and 'traditionalists'—Belgium.

But global co-authorship blurs national strokes. A comparison on a specific topic, such as information behavior, showed that the Jensen-Shannon distance (a measure from information theory) between a country's profile and the global benchmark is steadily shrinking. Differences, confirmed by the Kolmogorov-Smirnov test, a classic in probability theory, are fading. The share of papers using multiple methods rose from 15% to nearly 50%, and the 'fast-growing' countries contributed the most. Handwriting is becoming universal: researchers are increasingly writing in a common style.

Methodological graphology offers a window into the kitchen of science. Understanding where strengths lie, in experiments or theory, allows targeted selection of international partners. The next step is automatic classification of methods across languages using self-supervised learning and Large Language Models, which can not only recognize but also explain methodological choices. Maybe one day we'll see real-time scientometrics — a graphology of science where each new study immediately lands on the global map of handwriting. And if today's paper is still someone's autograph, tomorrow's resembles an inscription made by many hands, where individual traits dissolve into common form.