3 ms·

Nope. What you or your algorithm may find 'toxic', I might find interesting or useful.

• andreyk 6y ago
  Thanks for feedback! We are building it on top of a model trained with a large dataset of edits/comments editors on wikipedia found toxic or insulting. You can see more about the model here : https://github.com/tensorflow/tfjs-models/tree/master/toxicity https://github.com/tensorflow/tfjs-models/tree/master/toxici... So we are not making up how to classify toxicity really, moreso we are making the model usable via a browser extension.
  • mikecoles 6y ago
    I'm still not sure. I do wish you well in expanding the field. It's a difficult problem. Have you thought about personalizing the model with input from what a user has liked/upvoted on various forums? Even with that, it encourages bubbles and echo chambers though. I'm from a group that thrives on busting of um.. chops. If someone isn't jabbing you, that's when you should be concerned. Don't tell the governor, but a group was hanging out tonight and hairstyles, music choices, fatness, and politics were all fair game. The same goes for online groups, family, and work. Facebook and Twitter have tried this with the result being a total failure. Again, it's a very difficult problem and if your team is able to pull it off, the product is going to be worth more than a browser plugin. Best wishes and much respect for taking on a difficult task.
    • andreyk 6y ago
      Yes, personalization would be cool as a step after MVP!