About MapHub
MapHub ranks each day's new arXiv papers in astrophysics (astro-ph) and artificial intelligence (cs.AI) for a small community of researchers and groups (five as of October 2026), and shows whose research each paper matches best: a starting point for finding someone to ask about it.
Unlike tools that rank papers for one reader in private, MapHub scores every paper for every participant and shows the scores together; its profiles are portfolios that participants write and correct themselves; and its scores are public, with each table's commit recording the portfolio version, settings and code behind it. On a typical day so far, about 6% of the new astro-ph papers score 30 or more for each participant, on average, a handful worth reading out of some ninety.
- Rows are the papers announced that day; columns are participants, shown by pseudonym.
- Each cell is a relevance score from 0 to 100. A language model (Claude) assigns it by reading the paper's title, authors and abstract against the participant's research portfolio: 100 = they would have written it, 80 = coauthored it, 60 = would cite it, 30 = may broaden their view, 0 = worth a glance at the title, but not the abstract.
- Select a participant (column name) to see the papers most relevant to them, and the participants with the most similar interests. Select a paper (its title) to see who it matters to most: a starting point for finding someone to ask about it. Similarity, r, is the correlation of that day's scores.
- Top papers lists one participant's highest-scored papers over a range of listing dates (by default the past month, top 50), optionally with other participants' scores for the same papers beside; each listing date opens that day's heatmap.
- Hover over a cell for its score, or over an arXiv number for the paper's link. The arrows, the date box and the calendar move between days, keeping a selected participant; in the calendar, days with a table are highlighted, and weekends and holidays have no listing.
Participants are researchers and research groups who have each provided a short research portfolio describing their work; the scores come from these portfolios. Columns carry pseudonyms rather than names. Scores are public and stay in the tables after a participant pauses or leaves; their column can then be shown greyed out or hidden from this page. Pseudonyms are not anonymity, and portfolios are sent to the language model's provider: see what participants agree to. If you would like to join, please email kailicao@umich.edu.
MapHub is a team project (a “carpentry”) of the Eric and Wendy Schmidt AI in Science Postdoctoral Fellowship at the Michigan Institute for Data and AI in Society (MIDAS), led by Kaili Cao with teammates Kyle Heyblom, Eva Albalghiti and Dimitrios Simatos. Its code and daily tables are public on GitHub; titles and authors come from DataCite, which holds arXiv's DOI records.
MapHub is under development. Its scores come from a language model and can be wrong: treat them as pointers to papers and people, not as judgments of anyone's work. Questions and suggestions are welcome: kailicao@umich.edu.
Loading…