Verified, then logged.
Sun 22 Nov 2026 · 00:24 KST / 15:24 UTC한국어 | ENSearch /

Breaking01 / AI

MapHub reports results for personalised paper selection

Sample plates, a pipette and a laptop sit on a sunlit genomics lab bench while a researcher in a white coat, seen from behind, loads samples into an analysis instrument. AI-generated image

MapHub's authors reported in a preprint published on 10 Nov that its paper selection captured about 70% of their highest-rated papers (n undisclosed, pooled authors' blind ratings); the paper is not peer-reviewed, and its abstract reports no independent replication results.

According to the abstract of MapHub: Daily Relevance Heatmaps of New arXiv Preprints for a Research Community, the system assesses each day's new submissions to arXiv, a repository of research preprints, for individual research-community participants. A large language model compares papers with portfolios describing each participant's research interests.

The authors report that MapHub publishes relevance scores as a heatmap, a display that uses colour to represent values. It is designed to show whose research each paper relates to.

In the authors' evaluation, the highest-scoring 10% of papers contained about 70% of the raters' own top 5% (n undisclosed, pooled authors' blind ratings). The abstract identifies the authors as the evaluators, so the reported selection performance reflects their own judgements of relevance.

The abstract identifies text matching as the comparison baseline. MapHub instead uses a language model to read participant-specific portfolios and assess papers' relevance to them. The abstract does not disclose the number of evaluators or the detailed versions of the models used.