A Dynamical Cartography of the Epistemic Diffusion of Artificial Intelligence in Neuroscience

This repository includes the original codes related to the study provided in the following paper:

• Fontaine, S. (2025). A Dynamical Cartography of the Epistemic Diffusion of Artificial Intelligence in Neuroscience. Available at: https://arxiv.org/abs/2507.01651.

This work relies on the scientometric dataset available on the following Zenodo repository: ... .

This dataset includes neuroscience papers published in journals both referenced by the Web of Science and SCImago Journal Rank, which has been extracted from a datadump of the Microsoft Academic Knowledge Graph (version of 2020-05-29 available at https://doi.org/10.5281/zenodo.3936556).

Each notebook is named with a number indicating the chronological order of execution:

1. 1 - knowledgeMap_SPECTER+UMAP.ipynb: Pipeline dedicated to the pre-processing of the titles and abstracts of the neuroscientific papers included in our database, to the lexical embedding of the latter with the SPECTER model (https://github.com/allenai/specter), and to the dimensional reduction of the obtained embedding space with the UMAP method, which transforms the original 768-dimensional space returned by SPECTER into a 2-dimensional latent one. This notebook has generated some transformed and/or filtered datasets, which were uploaded to the Zenodo repository mentioned below. The map shown in the notebook results from one realization on a small sample. The map shown in the aforementioned article is also the result of executing this notebook on the entire dataset.
2. 2 - knowledgeMapPlot_ClusterConceptNetworkBuilding: Plots the reduced knowledge landscape in two dimensions and clusters the points in epistemic regions. The final section builds networks of MAG fields of study associated with the papers contained in each of these clusters.
3. 2.1 - knowledgeMap_robustness_analysis: Performs a robustness analysis of the 2D knowledge map produced in the previous notebook. This notebook assesses how well the conceptual neighborhoods are conserved for each paper before and after the UMAP dimensional reduction. We also provide a comparative analysis of the MAG fields of study and the OpenAlex topics and keywords used to characterize the knowledge map clusters.
4. 3 - conceptNetwork_coreness-analysis: Study of the coreness of AI-related concepts on the semantic networks of each knowledge cluster.
5. 4 - knowledgeMap_diffusion_analysis: Study of the citation network between the articles plotted in the 2D knowledge map, which introduces the radius of gyration to assess the epistemic coverage of a given paper, particularly those related to AI.

Throughout the codes, some paths toward directories of inputs' importation and outputs' saving (notably for data and figures that are heavy or obtained with long-lasting execution) need to be updated at the discretion of the user. The same goes from the installation of some Python packages (use pip install package for this purpose).

Python version of execution's environment: 3.8.10.