The project is a clone of "NEP: New Economics Papers", available at
http://nep.repec.org. Since 1998, NEP submits new working papers in
the RePEc digital library to human selectors, who select papers by
subject for inclusion in specialized reports. The selection is
manual. Since 2004, we use machine learning software to help selectors
to find the relevant papers. The median weekly time a selector spends
on a report issue is only 10 minutes.

We want to produce a similar system for the biomedical sciences based
on PubMed.  The PubMed metadata is of better quality than RePEc
data. The use of specialized jargon is more intensive. Therefore
machine learning will produce better results. However, the challenge
is the scale of PubMed. In RePEc, with an average of 700 papers an
issue, we can give a founding selector a completely unsorted issue to
start with. PubMed is too large to do that. Thus we need an additional
workflow for the composition of the initial report issue. This will
ask rookie selectors to give us examples. We then submit a subset of
the weekly additions to them that we think is close to the
examples.