Show HN: Taggy – Open-source package to auto-tag input/text

35 points by turbosepp 2 years ago | 12 comments
Hello!

I'm building a typescript based frontend package to tag/categorize any form of textual input.

Straight to the demo -> https://open-taggy.github.io/demo/

Project focus:

- lightweight (frontend package without any backend server stuff)

- easy integration (currently npm/vanilla, frameworks to come)

- free and open source

It works on the basis of a pre-defined list of trigger words (glossary). With this approach there is no need for any external services or heavy backend/ML-things.

  • _boffin_ 2 years ago
    Been working on something similar since when I take notes, I don’t actually want to think about tags, but rather have them be autogenerated and or suggested without losing my current context.

    Looking forward to seeing how you implemented this when I get home

    • geoelectric 2 years ago
      That's the use case I came in for. I'd love to use something like this with Obsidian or Logseq to autotag entries.

      Since mine tend to be domain-specific though, I do wonder if I wouldn't need either a more generalized model or a very specialized keyword list.

      • turbosepp 2 years ago
        Very interesting, thanks for sharing. I wonder about the integration to those. I took a quick look, Obsidian has a plugin system, so it needs to fit that I suppose. Not sure about Logseq right now...

        "Very domain specific" sounds like the need for a very specialized list. But it's hard to tell without knowing it. With the current approach it seems like some domains/use-cases work really over others (seemingly the "broader ones").

      • turbosepp 2 years ago
        That's an interesting use-case. I currently try to find out about those, thank you for that. ...would be great, if you have more feedback for me, especially as it's still work in progress :)
      • kris_wayton 2 years ago
        Looks like we've overwhelmed it? I'm seeing only "no matching tag found". Also seeing "Uncaught ReferenceError: example1 is not defined " in the JS console.
        • turbosepp 2 years ago
          Hopefully not :) ...It's still work in progress and I'm very thankful for these kind of error-/bug reports. Can you tell me a bit more please?

          The error-message indicates that this happens when you try the (online/linked above) demo-page, right? Hmm... what browser (and version) do you use?

          • turbosepp 2 years ago
            I think I already found it. Please try again :)
        • pimlottc 2 years ago
          One of the sample gifs shows a text about returning a product being tagged “RETURNMENT”… that’s not really a word anyone uses in that sense.
          • turbosepp 2 years ago
            Oh.. I'm not a native speaker and obviously used that wrong. Will change that, thank you!
          • KRAKRISMOTT 2 years ago
            Why do you not want to use a LDA model or other similarly established methods?
            • turbosepp 2 years ago
              I have to take a deeper look at the LDA model method first so I can't say too much about this right now. But one key focus in this project was explicitly NOT to use methods that need "more processing power" or a backend-service, e.g. and instead use a seemingly "trivial" method like looking for occurencies matched with a keyword-list. But maybe LDA can also be implemented in a frontend package? I will take a look, thanks for the hint. And it would also be helpful for me if you can give me some info about the other "similarly established methods". Thank you!
              • KRAKRISMOTT 2 years ago
                The keyword you are looking for is "topic modeling". For SOTA (state of the art) methods, look on arxiv.