5 ms·

The gallery is intentionally plain HTML; Google appears to be indexing all public documents on PDFy correctly (both viewer pages and actual PDF files). Unlisted

The gallery is intentionally plain HTML; Google appears to be indexing all public documents on PDFy correctly (both viewer pages and actual PDF files). Unlisted documents get a noindex tag. That said, I really need to write some code to extract the document metadata from the PDF files and display it on page; right now, the only thing that search engines (and the site itself) have to go off is the filename, which is far from optimal.

walterbell 12y ago
Ideally, metadata extraction would be done on upload and presented to the user for optional manual correction. This would be a major contribution to findability, because PDFs often have incorrect metadata (try searching for anything on archive.org) or the person uploading may have metadata relevant to a use case that is unforeseen by the document creator. In fact, the same file could be uploaded multiple times with different metadata. There's room here for experimentation, e.g. publishing content hashes and linking "duplicates" that have different metadata. If PDF metadata can be published in a structured format, it should be then be possible for Calibre or Docear / Zotero to import the PDF + JSON metadata directly into the document database.
- joepie91_ 12y ago
  The problem is that I don't want to add any more roadblocks to the upload process - it should be as 'instant' as possible. Even the 'public' vs. 'unlisted' selection took quite some consideration. Perhaps crowdsourcing metadata might be an idea - but that involves quite a bit more complexity, implementation-wise.
  - walterbell 12y ago
    How about an "edit metadata" button linked to the submitter's session cookie, which is only active for a few minutes, similar to HN post editing? For those who don't care, upload process stays the same. Those who want to edit have the option, within a few mins.
    - ratpik 12y ago
      Why not build a re-CAPTCHA type service around crowd sourcing PDF metadata. Read a PDF while you wait for something to happen. Too idealistic?
- fiatjaf 12y ago
  Linking hashes would free a lot of space, and letting each user place its own piece of metadata on each hash it wants would solve all the metadata and user trustworthiness problems. Google would do the rest.

walterbell 12y ago

Ideally, metadata extraction would be done on upload and presented to the user for optional manual correction. This would be a major contribution to findability, because PDFs often have incorrect metadata (try searching for anything on archive.org) or the person uploading may have metadata relevant to a use case that is unforeseen by the document creator. In fact, the same file could be uploaded multiple times with different metadata. There's room here for experimentation, e.g. publishing content hashes and linking "duplicates" that have different metadata. If PDF metadata can be published in a structured format, it should be then be possible for Calibre or Docear / Zotero to import the PDF + JSON metadata directly into the document database.

joepie91_ 12y ago
The problem is that I don't want to add any more roadblocks to the upload process - it should be as 'instant' as possible. Even the 'public' vs. 'unlisted' selection took quite some consideration. Perhaps crowdsourcing metadata might be an idea - but that involves quite a bit more complexity, implementation-wise.
- walterbell 12y ago
  How about an "edit metadata" button linked to the submitter's session cookie, which is only active for a few minutes, similar to HN post editing? For those who don't care, upload process stays the same. Those who want to edit have the option, within a few mins.
  - ratpik 12y ago
    Why not build a re-CAPTCHA type service around crowd sourcing PDF metadata. Read a PDF while you wait for something to happen. Too idealistic?
fiatjaf 12y ago
Linking hashes would free a lot of space, and letting each user place its own piece of metadata on each hash it wants would solve all the metadata and user trustworthiness problems. Google would do the rest.