Follow the paper. Read the conversation.

Hacker News commenters have been linking arXiv papers since the site opened in 2007. HackerPapers searches the Algolia HN index for arXiv, alphaXiv, Hugging Face Papers, ar5iv and legacy arXiv links and collects papers from comment links, submission URLs and links in submission text. Search includes every published paper; the main ranking uses the number of different commenters who linked it. The searches cover each source’s available history and continue with daily updates. This still cannot account for unindexed or deleted comments, unrecognized links or papers outside arXiv. The coverage dashboard shows completed searches, outstanding metadata and a lookup for missing papers. Each paper page links to its original submissions and any collected comments that mention it, so you can read the source conversations. It is the paper counterpart to hackertube.org (videos) and hackerbooks.org (books).

What counts as a mention

A mention is one Hacker News comment that contains a link to a paper on arxiv.org (abstract, PDF, HTML or listing pages, including ar5iv and export mirrors), or to the same paper on alphaXiv, Hugging Face Papers or SciRate, or the text arXiv:1706.03762. Version suffixes like v2 are ignored, so all versions of a paper count as one. A comment that links the same paper twice counts once.

A paper linked in a submission URL or its text is included as soon as its metadata is verified, even when no collected comment links to it. These papers appear in search and HN submissions. Submissions do not increase the distinct-commenter ranking. The reply count of a submission describes its whole HN discussion; it is not a count of comments linking the paper.

Why different people, not comments

The main ranking counts distinct commenters who linked a paper. One enthusiast posting the same paper forty times is one person. Forty people posting it once is a signal that the paper has reached a community. Ties are broken by the number of threads, then comments, then the most recent mention.

The main collection requires at least two different commenters. Early discoveries separately lists papers with verified metadata and one named commenter. A single mention is an opportunity to explore, not a quality rating. References awaiting metadata remain visible in the coverage lookup and data export.

The views

• Top is the all-time ranking. "For you" reweights it towards the arXiv categories of papers you have opened or saved; that preference is computed in your browser from localStorage and never sent anywhere.
• Discover offers seven- and thirty-day attention, first observed mentions, older papers discussed again after a quiet month, and papers discussed in at least five different years. “Since your last visit” uses a timestamp stored only in your browser. All dates refer to the collected sources.
• Early contains verified papers mentioned by exactly one named commenter, separately from the main ranking.
• Submissions lists verified papers linked in HN submissions, newest submission first. It includes papers with no linking comments and can overlap with Top and Early. An [S] label identifies papers outside the commenter ranking.
• Rising ranks by distinct people in the last 365 days, then the last 90.
• Random shuffles the whole list with a seed derived from today's date, so everyone sees the same random order on a given day. "Shuffle again" gives you a private one.
• Fields groups arXiv categories into fifteen broad fields. A paper is filed under its primary category as assigned on arXiv.
• Years is the year of the paper's first arXiv version, not the year Hacker News talked about it. Shor's algorithm is from 1995 even though nobody on HN linked it before 2007.
• Saved is a reading list kept in your browser. Export it as BibTeX or JSON, and import a JSON backup to merge saved papers from another browser. Add a reading status, private tags and notes; include them in your JSON backup. No account is needed; your reading list stays on your device. Unsaving a paper preserves its notes, which you can reopen on its paper page.

The small bar chart on each row is the paper's mention curve: Hacker News comments per year from 2007 to today, with the peak year in orange. The chart on the front page is the same curve for the whole corpus.

Find your next paper

Search by title, author surname or arXiv id across all published papers by default, including Early discoveries and submissions. The scope selector can narrow the search to a specific collection. Enable “Search abstracts” to include full author names and original abstracts; the additional search index downloads only when requested. Combine a category, publication year and minimum number of commenters, then sort the results. The URL keeps these filters, so you can bookmark or share a particular view. Choose “Least mentioned” to explore papers beyond the familiar names, or “Past 90 days” for recent attention. Preview opens the original abstract and recent HN excerpts beside your results; it does not generate a summary or judge whether a comment is positive.

Follow a subject or paper

Subscribe to the recent discussions RSS feed, choose a subject feed, or follow the feed on any paper page. Feed entries link to original HN comments or submissions and identify the source type. The submissions feed follows new paper submissions separately. Feed updates follow the published data snapshot.

Data freshness

Comments and stories come from the Algolia Hacker News Search API, paper metadata (title, authors, abstract, categories) from the arXiv API. An automated update is scheduled daily from the maintainer’s computer; if it is offline or an upstream service is unavailable, the previous build stays available. Check the build date for freshness; the footer of every page says when. The "last 365 days" and "last 90 days" windows are relative to that build time. Recent comments can take a little while to appear in Algolia's index, so the latest day is always a little underreported.

Papers with unavailable metadata or a withdrawal notice recognized by the pipeline are left out. Author names and titles are shown exactly as arXiv returns them.

Download the data

Everything on this site is derived from three files, regenerated with each build. They are free to use; please credit Hacker News (Y Combinator), arXiv and this site.

• papers.csvone row per recognized paper ID (metadata may be missing): id, title, authors, year, category, people, mentions, threads, first and last mention, submission count and dates
• mentions.csvone row per mentioning comment: arXiv id, comment id, author, date, story id
• papers-full.json.gzevery paper ID recognized in the collected sources, including single-person references, with metadata and the ids of the comments and submissions that mention it

The site also reads papers.json (main ranking), early.json, submissions.json and stats.json. The submissions list overlaps with the two commenter collections; deduplicate by paper slug when combining them.

Keyboard shortcuts

Credits

Hacker News data via the Algolia HN Search API. Paper metadata via the arXiv API; thank you to arXiv for use of its open access interoperability. Papers are set in Latin Modern, the OpenType successor of Knuth's Computer Modern, under the GUST Font License. The rest is Verdana, like Hacker News.

HackerPapers is not affiliated with Y Combinator, Hacker News or arXiv. It is built and run by Kai Pfister (kaipfister.de). No analytics, no cookies, no third-party requests.