1

For a class project in machine learning, I am considering building a set of predictors where the feature set includes information about the topic, keywords in the title, the authors, the date, etc., and the outputs are probability of getting published, expected citation impact, etc.

I would like to choose a single journal such as Nature for example. Will it be possible to collect the necessary data to accomplish this? Anyone have tips on where I can find statistical data about particular research journals?

Improve this question
1
  • Would it be sufficient to train against accepted papers without including submitted papers?
    – Andrew Drozdov
    Oct 15, 2015 at 17:16

2 Answers 2

Reset to default
1

IMHO, this is unlikely to fly. Some journals might give you statistics on the number of submitted and accepted publications. But submission to a journal is typically confidential and as an author, I would be pretty upset if Nature revealed this information to someone else.

Where you might be able to do similar analysis is a site like arXiv.org. It's used heavily in physics in other fields as a preprint service, and there are usually notes where and when a submission is accepted to a journal. Keywords are also included. The difference is that not everyone submitting to Phys Rev necessarily uses arXiv.org.

Improve this answer
1

If you can deal with metadata only (i.e. without publication record), arXiv is a wonderful data to start with. They have an AIP, but what you want to use is the Open Archive Initiative API, see my post (and answer): Getting a dump of arXiv metadata. In particular, you get self-reported publication data (journal, DOI).

For an even more relevant dataset, American Physics Society (Physical Review) has an established route for sharing data: http://journals.aps.org/datasets:

[...] Requests will be quickly reviewed and, if approved, the data will be made available for download after accepting the terms and conditions below. [...] The corpus of Physical Review Letters, Physical Review, and Reviews of Modern Physics is comprised of over 450,000 articles and dates back to 1893. [...]

1) Citing article pairs: This data set consists of pairs of APS articles that cite each other. For instance, if article A cites article B, there will be an entry in the data set consisting of the pair of DOIs for A and B. This data set will be formatted as a comma-separated values (CSV) file consisting of the DOI pairs, one pair per line.

2) Article metadata: This data set consists of the basic metadata of all APS journal articles. The metadata provided includes the following fields: DOI, journal, volume, issue, first page and last page OR article id and number of pages, title, authors, affiliations, publication history, PACS codes, table of contents heading, article type, and copyright information.

I don't know if a class project counts, but it may be worth trying.

Improve this answer

You must log in to answer this question.