sonos / nlu-benchmark Goto Github PK

Hey, you said "Note that None intents were ignored when computing the metrics, as in the original article."

Can you please refer the exact place where this None intent was ignored in the original article? Tried but couldn't find.

Thanks

Greg

Hi

I am referencing the https://www.slideshare.net/KonstantinSavenkov/nlu-intent-detection-benchmark-by-intento-august-2017 , in this slide share on side has accuracy for False Positive.

I didn't understand correctly how You came across to that figure. Can you please explain how this test has done? and also What is the data set used for this False positive test. It has mentioned that NLU 2016 samples are used for False positive , Though intent names are different in NLU 2016 , some of the quires are matching with quires in NLU 2017,which we have used for training. So I am bit confused on this. Can you explain the same and also please provide link for the data set used for this test if it is publicly available.

README points to https://sdk.snips.ai/
But there is no listener on 443.

Hello, the chatbot results located at https://github.com/snipsco/nlu-benchmark/blob/master/2018-01-Braum-et-al-extension/results/chatbot/snips_metrics.json are wrong, because this file includes the testing against the training data. Test data for DepartureTime intent is 35 sentences and the result contains 80, and FindConnection should be 71 total sentences, and the result contains 130.

In order to benchmark a SLU?

Hi there! I implemented an intent classifier using Convolutional Neural Networks by using this repository's data. Here's a link to it, repo I used it just for the demonstration purpose and hope it is fine with you!

The example starting at line 9085 has errors (lexical) and also has some weird characters that can't be decoded, generating weird errors when loading or manipulating. Please have a look at that.
See below:

{
  "data": [
    {
      "text": "I want toi hear some "
    },
    {
      "text": "Pop Punk Perfection ������",
      "entity": "playlist"
    },
    {
      "text": " off of "
    },
    {
      "text": "Deezer",
      "entity": "service"
    }
  ]
},