CognitiveExperimentation

Experimentation for harassment detection with Azure Cognitive Services.

There are two ways to try it out:

• A console app
• An API with a Vonage API integration (Messages)

Console Application

This application reads data from input.json, and outputs results into output.json.

How to setup the project

Set your Azure key and endpoint in appsettings.json.

{
  "CognitiveOptions": {
    "ApiKey": "...",
    "ApiEndpoint": "..."
  }
}

Add sentences you want to evaluate in input.json.

{
  "Values": [
    "Sentence 1",
    "Sentence 2",
    "Sentence 3"
  ]
}

Run the project

Simply run the project and observe the console output. You should see each sentence being evaluated by Cognitive Services.

When the process finished, the file output.json containing detailed results is created in \bin\Debug\net6.0.

API - Vonage integration with Messages

This is a local API working with Vonage Messages API.

Here's the workflow:

• The client sends an SMS to your Vonage number
• Our API receives the message content via a callback
• Our API sends the text to Azure Cognitive Services
• Our API formats the data
• Our API sends an SMS back to the client, with the classification details

How to setup the project

Fill your credentials in appsettings.json.

{
  "CognitiveOptions": {
    "ApiKey": "Your Azure Key",
    "ApiEndpoint": "Your Azure Endpoint",
    "VonageApplicationId": "Your Vonage ApplicationId"
  }
}

Fill your Vonage private key in PrivateKey.txt.

-----BEGIN PRIVATE KEY-----
...
-----END PRIVATE KEY-----

By then, you should be able to run the API locally with SwaggerUI.

How to expose your API with ngrok

By default, the API runs over http://localhost:5000

Run the following to expose this port through tunneling on ngrok:

ngrok http 5000

Your local APi now has a public url, like https://4a8698326dae.ngrok.app.

How to setup the Callback Url

On Vonage Dashboard, navigate to your application - make sure to have Messages enabled.

You need to set Messages inbound url with the following value: {ngrok-url}/harassment with ngrok-url being your public url.

Finally, you need to link a Vonage phone number to this Application.

Run the project

Simply run the project, and send a text message to your Vonage phone number.

You will receive a text message with the following structure:

Sexually explicit: 1%
Sexually suggestive: 25%
Offensive: 18%