Code and data network: datasets and repositories used together
Which datasets and which code, models and data repositories research papers use together, as a network: its groups, the nodes that bridge them, the strongest and the most distinctive pairs, and the flow from datasets to code. A node is a dataset that papers name in their title, abstract or full text (as on the dataset rankings), used by at least 10 papers, or a code or data repository or software package that papers link to in their Crossref records (as on /code), linked by at least 5. Two nodes are joined when at least 3 papers use both; a paper using more than 40 of them (a survey or a benchmark list) joins none. Groups are the communities the Louvain method finds (nodes more often used with each other than with the rest). Centrality is the betweenness: the share of shortest paths between other nodes that pass through a node, so a high value marks a link between groups. Bridging is the participation coefficient: how evenly a node's links spread over the groups (0: all in one group). Lift compares the papers using both with what the two nodes' sizes alone would give (1: no more than chance). Datasets are found by name and repositories by links, so both are lower bounds. Also as Markdown and JSON (key "network").
41,608 papers in the tracked journals use at least one of the 438 datasets and repositories in this network (350 datasets, 88 code and data repositories and packages); 15,324 use two or more, which make its 5,078 links.
The Louvain method finds 11 groups of three or more; the largest are around World Development…, Eurostat, International Energy… (115), S&P 500, Economic Policy…, Compustat (88) and ImageNet, CIFAR-10 / CIFAR-100, MNIST / Fashion-MNIST (87).
The most central nodes, which link otherwise separate groups, are Twitter / X, Facebook and Kaggle.
The dataset and code used together most often are MS COCO and ultralytics/ultralytics (22 papers).
The most distinctive pair is doParallel (R package) with foreach (R package): 9 papers, 3,467 times what chance would give.
A node is a dataset that papers name in their title, abstract or full text (as on the dataset rankings), used by at least 10 papers, or a code or data repository or software package that papers link to in their Crossref records (as on /code), linked by at least 5. Two nodes are joined when at least 3 papers use both; a paper using more than 40 of them (a survey or a benchmark list) joins none. Groups are the communities the Louvain method finds (nodes more often used with each other than with the rest). Centrality is the betweenness: the share of shortest paths between other nodes that pass through a node, so a high value marks a link between groups. Bridging is the participation coefficient: how evenly a node's links spread over the groups (0: all in one group). Lift compares the papers using both with what the two nodes' sizes alone would give (1: no more than chance). Datasets are found by name and repositories by links, so both are lower bounds.
Cite this page
Scholar4AI. (2026). Code and data network: datasets and repositories used together. Journal and Conference Research Rankings. Retrieved November 4, 2026, from https://scholar4ai.com/code/network
@misc{scholar4ai-code-network,
author = {{Scholar4AI}},
title = {Code and data network: datasets and repositories used together},
year = {2026},
howpublished = {\url{https://scholar4ai.com/code/network}},
urldate = {2026-11-04},
note = {Journal and Conference Research Rankings; data from OpenAlex (CC0); licensed CC BY-NC 4.0}
}
Free to reuse for non-commercial purposes with attribution (CC BY-NC 4.0).