Paper Firehose
Menu

About

Paper Firehose collects new research papers from journal and arXiv feeds and keeps the ones that matter to you.

Topics

A topic is one research interest. It has three parts:

How a run works

A run processes one topic in four steps. It usually takes a few minutes. Only one of your topics can run at a time.

  1. Filter. The topic's feeds are downloaded. A paper is kept when its title or its feed summary matches the filter keywords. Upper and lower case make no difference. Papers seen in an earlier run are skipped, so each run brings only what is new.
  2. Rank. Each kept paper gets a score between 0 and 1. The score says how close the title is in meaning to the ranking keywords. A language model compares them, so the exact words need not appear. A paper that contains a negative keyword loses 0.25. A paper by a preferred author gains 0.20.
  3. Abstracts. For papers that score at or above the topic's rank threshold (0.30 unless you change it), the full abstract is looked up in Crossref, Semantic Scholar, OpenAlex and PubMed. arXiv papers bring their abstract with the feed. The threshold removes no papers: lower-scoring ones stay in the results, without a looked-up abstract.
  4. Results. The topic's results page shows the papers of the latest run, at most the best 200, ordered by score or grouped by journal. Every match is also added to the topic's history, which you can search and filter.

Changing keywords

Because seen papers are skipped, new keywords only apply to papers that arrive later. To match the papers already in the feeds again, tick Reprocess when you start the run. The history is kept.

Email reports

In Settings you can choose topics, days and a time. On those days the pipeline runs for the chosen topics and a mail goes to the address you sign in with. It lists the papers matched since the previous report, at most the best 20 per topic. The first report covers the last 7 days. If nothing new matched, no mail is sent.

Built on paper-firehose

This site is built on top of paper-firehose, a Python package. Filtering, ranking and looking up abstracts is done by that package. The site adds accounts, the topic forms, the results pages and the scheduled mails around it.

You do not need this site to use it. Anyone can install the package and run their own paper firehose on their own computer, with their own feeds and topics: pip install paper-firehose The repository on GitHub explains how to set it up.

If you have an account here, Settings → Your data downloads your topics and papers as a folder that is ready to run with the package.