What is heating up on arXiv
Every cluster of arXiv, measured against arXiv's own growth.
arXiv publishes far more than it used to, so "more papers than last year" describes almost everything and distinguishes nothing. The interesting question is which subjects are taking a larger share of what gets published, and which are quietly giving theirs up. That is what the number below measures: a cluster's share of the last twelve complete months, divided by its share of every month since 1991. Above 1 it is gaining ground, below 1 it is losing it, and 1 means it is growing exactly as fast as arXiv.
Gaining share fastest
Losing share fastest
What this does not say
- Cooling is not shrinking. A cluster can publish more papers every year and still fall on this list, because arXiv grew faster. Supersymmetry breaking has not stopped; it is a smaller fraction of a much larger arXiv.
- Clusters are not disciplines. They come from the wording of titles and abstracts, so a field can be split across several clusters or share one with a neighbour. Read a cluster's papers before drawing conclusions about a field.
- New topics are handicapped. A subject younger than two years, or holding fewer than 200 papers in total, is left out entirely — there is no history to compare it against, and a handful of papers is not a trend.
- Naming is automatic. Cluster labels are written by a local language model from each cluster's papers. They are usually apt and occasionally clumsy.
- One snapshot, one harvest. The figures come from a single metadata harvest, stated below; the incomplete final month is excluded.
Check it yourself
The figures here come from the published dataset, which is free to download under CC-BY-4.0 — so you can recompute them yourself: for each cluster, take its papers in the last twelve complete months as a share of all papers in that period, and divide by its share over the whole archive. The map colours every cluster by the same number: open it or start at the overview.